<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Andrew on Medium]]></title>
        <description><![CDATA[Stories by Andrew on Medium]]></description>
        <link>https://medium.com/@andrewwongai?source=rss-f91d8ad9e8a2------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*O7dKm7niEbmYJFr-vwHIcQ.png</url>
            <title>Stories by Andrew on Medium</title>
            <link>https://medium.com/@andrewwongai?source=rss-f91d8ad9e8a2------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 21:25:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@andrewwongai/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Beginner’s guide to Stable Diffusion models and the ones you should know]]></title>
            <link>https://medium.com/@andrewwongai/beginners-guide-to-stable-diffusion-models-and-the-ones-you-should-know-d2ffd3ce992f?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/d2ffd3ce992f</guid>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Tue, 29 Nov 2022 13:15:21 GMT</pubDate>
            <atom:updated>2022-11-29T13:20:32.457Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Beginner&#39;s guide to Stable Diffusion models and the ones you should know" src="https://cdn-images-1.medium.com/max/800/0*T0kWpyDu3s7VDn7x.png" /></figure><p><strong>Models</strong>, sometimes called <strong>checkpoint files</strong>, are pre-trained Stable Diffusion weights intended for generating general or a particular genre of images.</p><p>This is part 3 of the beginner’s guide series.</p><p>Read part 1: <a href="https://stable-diffusion-art.com/beginners-guide/">Absolute beginner’s guide</a>.</p><p>Read part 2: <a href="https://stable-diffusion-art.com/inpainting_basics/">Inpainting</a>.</p><p>(This article was first published in <a href="https://stable-diffusion-art.com/">https://stable-diffusion-art.com/</a>)</p><p>What images a model can generate depends on the data used to trained them. A model won’t be able to generate a cat’s image if there’s never a cat in the training data. Likewise, if you only train a model with cat images, it will only generate cats.</p><p>We will go introduce what models are, some common ones (<a href="#Stable_diffusion_v14">v1.4</a>, <a href="#Stable_diffusion_v15">v1.5</a>, <a href="#F222">F222</a>, <a href="#Anything_V3">Anything V3</a>, <a href="#Open_Journey">Open Journey v4</a>), how to install, use and merge them.</p><h3>Fine-tuned models</h3><h4>What is fine tuning?</h4><p><strong>Fine tuning</strong> in a common technique in machine learning of taking a model that is trained on a wide dataset, and train a bit more on the narrow dataset that you are interested in.</p><p>A fine-tuned model will be biased towards generating images similar to your dataset, while maintaining the versatility of the original model.</p><h4>Why do people make them?</h4><p>Stable diffusion is great but is not good at everything. For example, it can and will generate anime style images with the keyword “anime” in prompt. But it could be difficult to generate images of a sub-genre of anime. Instead of tinkering with the prompt, you can fine tune the model with images of that sub-genre.</p><h4>How are they made?</h4><p>There are two main fine tuning methods: (1) Additional training and (2) Dreambooth. They both start with a base model like Stable Diffusion <a href="https://stable-diffusion-art.com/wp-admin/post.php?post=806&amp;action=edit#Stable_diffusion_v14">v1.4</a> or <a href="https://stable-diffusion-art.com/wp-admin/post.php?post=806&amp;action=edit#Stable_diffusion_v15">v1.5</a>.</p><p><strong>Additional training</strong> is achieved by training a base model with an additional dataset that you are interested in. For example, you can train Stable Diffusion v1.5 with an additional dataset of vintage cars to bias the aesthetic of cars towards the sub-genre.</p><p><a href="https://arxiv.org/abs/2208.12242"><strong>Dreambooth</strong></a>, originally developed by Google, is a technique to inject custom subjects into text-to-image models. It works with as few as 3–5 custom images. You can take a few pictures of yourself and use Dreambooth to put yourself into the model. A model trained with Dreambooth requires a special keyword to condition the model.</p><p>There’s another less popular fine tuning technique called <a href="https://textual-inversion.github.io/"><strong>textual inversion</strong></a>. The goal is similar to Dreambooth: Inject a custom subject into the model with only a few examples. A new keyword is created specifically for the new object. Only the text embedding network is fine tuned, while keeping the rest of the model unchanged. In layman’s term, it’s like using existing words to describe a new concept.</p><h3>Models</h3><p>There are hundreds of fine-tuned Stable Diffusion models and the number is increasing everyday. Below is a list of model that can be used for general purpose.</p><h4>Stable diffusion v1.4</h4><p><a href="https://huggingface.co/CompVis/stable-diffusion-v-1-4-original">Model Page</a></p><p><a href="https://huggingface.co/CompVis/stable-diffusion-v-1-4-original/resolve/main/sd-v1-4.ckpt">Download link</a></p><p>Released in August 2022 by Stability AI, v1.4 model is considered to be the first publicly available Stable Diffusion model.</p><p>You can treat v1.4 as a general-purpose model. Most of the time, it is enough to use it as is unless you are really picky about certain styles.</p><h4>Stable diffusion v1.5</h4><p><a href="https://huggingface.co/runwayml/stable-diffusion-v1-5">Model Page</a></p><p><a href="https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.ckpt">Download link</a></p><p>v1.5 is released in Oct 2022 by Runway ML, a partner of Stability AI. The model is based on v1.2 with further trainings.</p><p>The <a href="https://huggingface.co/runwayml/stable-diffusion-v1-5">model page</a> does not mention what the improvement is. It produces slightly different results compared to v1.4 but it is unclear if they are better.</p><p>Like v1.4, you can treat v1.5 as a general-purpose model.</p><p>In my experience, v1.5 is a fine choice as the initial model and can be used interchangeably with v1.4.</p><h4>F222</h4><p><a href="https://ai.zeipher.com/">Model Page</a></p><p><a href="https://mega.nz/file/0c8hkagB#ER-Lz6Wwb3y2BXko4ZzA0kfusnGYk-EYsz3E3vugfqo">Download link</a></p><p><strong>F222</strong> is trained originally for generating nudes but people found it useful in generating beautiful female portraits with correct body part relations. Interestingly, contrary to what you might think, it’s quite good at generating aesthetically pleasing clothings.</p><p>F222 is a special purpose model but it is quite useful for portraits. It has a high tendency of generating nudes but it can be suppressed by keywords like “dress”.</p><p>Read <a href="https://ai.zeipher.com/f.html">recommended keyword</a> list.</p><h4>Anything V3</h4><p><a href="https://huggingface.co/Linaqruf/anything-v3.0">Model Page</a></p><p><a href="https://huggingface.co/Linaqruf/anything-v3.0/resolve/main/Anything-V3.0-pruned.ckpt">Download Link</a></p><p><strong>Anything V3</strong> is a special-purpose model trained to produce high-quality anime-style images. You can use <a href="https://danbooru.donmai.us/tags">danbooru tags</a> (like 1girl, white hair) in text prompt.</p><p>It’s useful for casting celebrities to amine style, which can then be blended seamlessly with illustrative elements.</p><p>One drawback (as least to me) is that it produces females with disproportional body shapes. I like to tone it down with F222.</p><h4>Open Journey</h4><p><a href="https://huggingface.co/prompthero/openjourney">Model Page</a></p><p><a href="https://huggingface.co/prompthero/openjourney/resolve/main/mdjrny-v4.ckpt">Download link</a></p><p>Open Journey is a model fine-tuned with images generated by <a href="https://www.midjourney.com/">Mid Journey v4</a>. It has a different aesthetic and is a good general purpose model.</p><h4>Model comparison</h4><p>Here’s a comparison of these models with the same prompt and seed. All but Anything v3 generate realistic images but with different aesthetics.</p><figure><img alt="Compare commonly used models." src="https://cdn-images-1.medium.com/max/1024/0*GvuijI7Vr5cWBE68.png" /></figure><h4>Images generated with the same seed and steps.<br><br>Other models</h4><p>There are hundreds of Stable Diffusion models available. Many of them are special-purpose models designed to generate a particular style. Some notable ones are:</p><p>- <a href="https://huggingface.co/hakurei/waifu-diffusion">waifu-diffusion:</a> Japanese anime style<br><br>- <a href="https://huggingface.co/nitrosocke/Arcane-Diffusion">Arcane-Diffusion</a>: TV show Arcane style<br><br>- <a href="https://huggingface.co/nousr/robo-diffusion">robo-diffusion</a>: Robot images<br><br>- <a href="https://huggingface.co/nitrosocke/mo-di-diffusion">mo-di-diffusion</a>: Modern Disney style</p><p>You can find more models <a href="https://huggingface.co/models?other=stable-diffusion">here</a>.</p><h3>How to install and use a model</h3><p>To install a model in AUTOMATIC1111 GUI, download and place the checkpoint (.ckpt) file in the following folder</p><p>stable-diffusion-webui/models/Stable-diffusion/</p><p>Press <strong>reload</strong> button next to the checkpoint drop box</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hrwsP21ANzEkUDrS.png" /></figure><p>You should see the checkpoint file you just put in available for selection. Select the new checkpoint file to use the model.</p><p>If you are new to AUTOMATIC1111 GUI, some models are preloaded in Colab notebook included in the <a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start?layout=profile">Quick Start Guide</a>.</p><h3>Merging two models</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UE_w7z292H0xDlJg.png" /></figure><p>Settings for merging two models.</p><p>To merge two models using AUTOMATIC1111 GUI, go to the <strong>Checkpoint Merger</strong> tab and select the two models you want to merge in <strong>Primary model (A)</strong> and <strong>Secondary model (B)</strong>.</p><p>Adjust the multiplier (M) to adjust the relative weight of the two models. Setting it to 0.5 would merge the two models with equal importance.</p><p>After pressing <strong>Run</strong>, the new merged model will be available for use.</p><p>Example of a merged model</p><p>Here are sample images from merging <a href="#F222">F222</a> and <a href="#Anything_V3">Anything V3</a> with equal weight (0.5):</p><figure><img alt="comparison between merged models" src="https://cdn-images-1.medium.com/max/1024/0*B2FJH2iiy_pUKJqn.png" /></figure><p>Compare F222, Anything V3 and Merged (50% each)</p><p>The merged model sits in between the realistic F222 and the anime Anything V3 styles. It is a very good model for generating illustration arts with human figures.</p><h3>Summary</h3><p>In this article, I have introduced what Stable Diffusion models are, how they are made, a few common ones, and how to merge them. Using models can make your life easier when you have a specific style in mind.</p><p>This is part 3 of the beginner’s guide series.</p><p>Read part 1: <a href="https://stable-diffusion-art.com/beginners-guide/">Absolute beginner’s guide</a>.</p><p>Read part 2: <a href="https://stable-diffusion-art.com/inpainting_basics/">Inpainting</a>.</p><p>Visit to support this free blog:</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_prompt_generator?layout=profile">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li></ul><p>Follow me on <a href="https://twitter.com/stable_diff_art">Twitter</a> &amp; <a href="https://www.instagram.com/stable.diffusion.art/">instagram</a> to receive content like this.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d2ffd3ce992f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to stylize images using Stable Diffusion AI]]></title>
            <link>https://medium.com/@andrewwongai/how-to-stylize-images-using-stable-diffusion-ai-3ddf776e2f03?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/3ddf776e2f03</guid>
            <category><![CDATA[img2img]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[image-style-transfer]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Tue, 22 Nov 2022 03:39:50 GMT</pubDate>
            <atom:updated>2022-11-29T13:21:57.284Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="How to stylize images using Stable Diffusion AI" src="https://cdn-images-1.medium.com/max/800/0*Xz83y7XRuIwFkjc6.png" /></figure><p>Image stylization is a process of transforming the style of an image to a new one. For example, you may want to create a pencil drawing from a photo you have taken.</p><p><em>(This article was first published in </em><a href="https://stable-diffusion-art.com/"><em>https://stable-diffusion-art.com/</em></a><em>)</em></p><p>There has been a long line of research in <strong>style transfer</strong>, starting with <a href="https://research.adobe.com/news/image-stylization-history-and-future-part-2/">hand-crafted algorithms</a> to the state-of-the-art <a href="https://research.adobe.com/news/image-stylization-history-and-future-part-3/">neural-network based methods</a>.</p><p>In this article, I will explore image stylization using Stable Diffusion, a text-to-image model trained with millions of image-text pairs.</p><p>While existing methods have already created <a href="https://research.adobe.com/news/puppetron-your-face-as-a-stylized-work-of-art/">impressive results</a>, the benefit of using text-to-image model is that all styles are already embedded in the model. No training is required. You will only need to specify the style you want with the <a href="https://stable-diffusion-art.com/beginners-guide/">text prompt</a>. Lastly, if you are not happy with the result of the embedded model, you can further fine-tune and create your own model.</p><h3>GUI</h3><p>I will show you how to do it with AUTOMATIC1111 GUI. See my <a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">quick start guide</a> for setting up in Google’s cloud server.</p><h3>Workflow for stylizing images</h3><h4>Basic idea</h4><p>The basic idea is to use <strong>img2img</strong> to modify an image with the new style specified in the <strong>text prompt</strong>.</p><p>Instead of using a random latent state, the original image was used to encode the initial latent state. With a small amount of randomness added, which is controlled by the <strong>denoising strength</strong>, the model constructs a new image according to the text prompt.</p><p>So if you include <strong>pencil drawing</strong> in your text prompt, the synthesized image will include visual elements of pencil drawing but the content of the image still follows the original one.</p><h4>Step-by-step workflow</h4><p>OK, enough of these theoretical stuffs! Here’s how you would actually do it.</p><p>I will use the following starting image generated using the <a href="https://stable-diffusion-art.com/victorian-girl/">Victorian girl prompt</a>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*PilI_5X4N0TCK5Bl.png" /><figcaption>The original image to be stylized.</figcaption></figure><p>In AUTOMATIC1111 GUI, go to <strong>img2img</strong> tab and select the <strong>img2img</strong> sub tab. Upload the image to the img2img canvas.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cuIAfp3L_XuFNZhq.png" /></figure><p>Next you will need to give a prompt. <strong>The prompt should describes both the new style and the content of the original image. </strong>It does not need to be super detailed. Below is what I will use.</p><p>a _______ of a beautiful woman</p><p><strong>Insert the art style to the blank.</strong> (E.g. a charcoal sketch of a beautiful woman)</p><p>For stylizing images, you will be most tweaking two parameters: <a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#CFG_Scale"><strong>CFG scale</strong></a> and <a href="https://stable-diffusion-art.com/inpainting_basics/#Denoising_strength"><strong>denoising strength</strong></a><strong>.</strong></p><p><strong>RECAP: CFG scale</strong> controls how much the model should follow the prompt. The higher the more it follows. <strong>Denoising strength</strong> controls how much the image should change. The lower the less changes it will be.</p><p>If you want to be systematic, you can use the <strong>X/Y plot script</strong> to explore the interplay between CFG scale and Denoising strength. The image is specified by the <a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#Seed">seed value</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*nPTbaGJGcpFJvblC.png" /><figcaption>X/Y plot script settings.</figcaption></figure><p>Using the prompt:</p><p>A pencil sketch of a beautiful woman</p><p>Below is what you will get.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xGvJ49Y4zdKm86bu.png" /><figcaption>X/Y Plot of CFG scale vs denoising strength</figcaption></figure><p>When denoising strength is high, the image changes too much and loses it original content. When it is low, not much is changed.</p><p>The higher the CFG scale, the stronger the style is.</p><p><strong>Euler</strong> sampling method with <strong>50 steps</strong> works for most situations.</p><p>As a rule of thumb, set CFG scale to a high value between 20–30, and change denoising strength between 0.2 and 0.6 to achieve a good stylization while preserving the content.</p><h4>Examples of stylized images</h4><p>All stylized images in this section is generated from the original image below with zero examples. Only text prompts are provided.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*h13K750M7jkTPKdy.png" /><figcaption>Original</figcaption></figure><p>Here are some stylized images. The prompt is</p><p>A ______ of a beautiful woman</p><p>The blank is filled in with the caption.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*lB6JBcwV6Gnx9gWY.png" /><figcaption>Charcoal</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*TO1WLSwO3teNUD3j.png" /><figcaption>Impressionist</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*mZatOH4bmIEFG7ZG.png" /><figcaption>Pop art</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*7-WF9PcIAxOMbonU.png" /><figcaption>Art Deco</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*HNwHQ-A2bpZvfVBp.png" /><figcaption>Photograph</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*gFir_aINZlDdf4Ug.png" /><figcaption>Art Nouveau</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*I_AmTdVPmKCwnAUn.png" /><figcaption>Cubism</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*SEoqLzmnDt3SVp5S.png" /><figcaption>John Sargent</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*VNIvmzT9s8bMhrcK.png" /><figcaption>Greg Rutkowski</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*pq34Ttky5AsptiPv.png" /><figcaption>3D character</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*qUFYYS2LFGztK59C.png" /><figcaption>Stone Sculpture</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*Zc-bcmwHG0Moewa4.png" /><figcaption>Van Gogh</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*pHn4FxBr0lsHPDeE.png" /><figcaption>Comic Strip</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*cv7a_FSyozvoeYlY.png" /><figcaption>Modern fashion</figcaption></figure><p>It is also possible to switch the model to a different ethnicity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*eZhsyzjnIwvF1VTH.png" /><figcaption>Japanese</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*YV6p6k_s3oJlaY5L.png" /><figcaption>African</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*EWawHkxzXFOnUhmA.png" /><figcaption>Hispanic</figcaption></figure><h3>Alternative method</h3><p>Adventurous readers can experiment with the so-called <a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#img2img-alternative-test">alternative img2img test</a> which can preserve the content of original image far better than the method above.</p><p>The idea is to use the original image to create the latent noise pattern so that that it contains information of the original content.</p><p>To use the alternative noise, select <strong>img2img alternative test</strong> in the script dropdown menu.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Gb_hIvu4ETbHcT96.png" /></figure><p>New users should go through the <a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#img2img-alternative-test">example</a> in the feature demo. Contrary to the demo, I found it useful to increase CFG scale so that the prompt is followed. Sometimes the <a href="https://stable-diffusion-art.com/fine-tune-your-ai-images-with-these-simple-prompting-techniques/#Adjust_keyword_strength_using_weight">weight</a> of the style keyword needs to be increased.</p><p>Below is an example of pop art style generated with alternative img2img.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*5ELE73Ygs3v1Qxz1.png" /><figcaption>Pop art style using alternative img2img.</figcaption></figure><p>Whether it is better would be a matter of taste but it definitely follows the original content closely.</p><p>Treat it as an tool in your toolbox.</p><h3>Summary</h3><p>I have showed you how easy it is to use Stable Diffusion to stylize images. All you need to do is to use <strong>img2img</strong> method, supply a <strong>prompt</strong>, dial up the <a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#CFG_Scale"><strong>CFG scale</strong></a>, and tweak the <a href="https://stable-diffusion-art.com/inpainting_basics/#Denoising_strength"><strong>denoising strength</strong></a>.</p><p>I hope this will inspire you to create some new artworks!</p><p><em>This blog post is originally published on </em><a href="http://stable-diffusion-art.com/"><em>stable-diffusion-art.com</em></a></p><p>Visit to support this free blog:</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_prompt_generator?layout=profile">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li></ul><p>Follow me on <a href="https://twitter.com/stable_diff_art">Twitter</a> &amp; <a href="https://www.instagram.com/stable.diffusion.art/">instagram</a> to receive content like this.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3ddf776e2f03" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beginner’s guide to inpainting (step-by-step examples)]]></title>
            <link>https://medium.com/@andrewwongai/beginners-guide-to-inpainting-step-by-step-examples-26445f31248a?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/26445f31248a</guid>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[ai-image-generator]]></category>
            <category><![CDATA[inpainting]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-art]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Thu, 17 Nov 2022 00:17:59 GMT</pubDate>
            <atom:updated>2022-11-17T00:37:04.403Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Beginner&#39;s guide to inpainting (step-by-step examples)" src="https://cdn-images-1.medium.com/max/1024/0*Cl9cQw1650Kj94ix.png" /></figure><p>No matter how good your prompt and model are, it is rare to get a perfect image in one shot.</p><p>Inpainting is a great way to fix small defects. In this post, I will go through a few basic examples to use <strong>inpainting</strong> for fixing defects.</p><p>If you are new to AI images, you may want to read the <a href="https://stable-diffusion-art.com/beginners-guide/">beginner’s guide</a> first.</p><h3>Image model and GUI</h3><p>We will use Stable Diffusion AI and AUTOMATIC1111 GUI. See my <a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">quick start guide</a> for setting up in Google’s cloud server.</p><h3>Basic inpainting settings</h3><p>In this section, I will show you step-by-step how to use inpainting to fix small defects.</p><p>I will use an original image from the <a href="https://stable-diffusion-art.com/lonely-palace/">Lonely Palace</a> prompt:</p><blockquote>[emma watson: amber heard: 0.5], (long hair:0.5), headLeaf, wearing stola, vast roman palace, large window, medieval renaissance palace, ((large room)), 4k, arstation, intricate, elegant, highly detailed</blockquote><p>(Detailed settings can be found <a href="https://stable-diffusion-art.com/lonely-palace/">here</a>.)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/0*j--WHAX6ziGb8U-3.png" /><figcaption>Original image</figcaption></figure><p>It’s a fine image but I would like to fix the following issues</p><ul><li>The face looks unnatural.</li><li>The right arm is missing.</li></ul><h3>Creating an inpaint mask</h3><p>In AUTOMATIC1111 GUI, Select the <strong>img2img</strong> tab and select the <strong>Inpaint</strong> sub-tab. Upload the image to the inpainting canvas.</p><figure><img alt="inpainting canvas" src="https://cdn-images-1.medium.com/max/1024/0*_9mpr5mZf47GyoJo.png" /></figure><p>We will inpaint both the right arm and the face at the same time. Using the paintbrush tool to create a mask to be inpainted.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*S0pUgzSFGlDEuTf6.png" /><figcaption>Create mask use the paintbrush tool.</figcaption></figure><h3>Settings for inpainting</h3><p>You can <strong>reuse the original prompt</strong> for fixing defects. This is akin to generating multiple images but you can select a particular area and settings.</p><p>The image size need to be adjusted to be the same as the original image. (704 x 512 in this case).</p><p>If your are inpainting faces, make sure to turn on <strong>restore faces</strong>. (You will also need to select and apply the <strong>face restoration model</strong> to be used in the <strong>Settings</strong> tab. <strong>CodeFormer</strong> is a good one.)</p><p>When inpainting small area with fine details, it is important to select <strong>inpaint in full resolution</strong>. It would upscale your inpainting area , perform inpainting, and scale it back down. The end result is</p><p>You can experiment with the <strong>Mask Content</strong> options. I found <strong>latent noise</strong> and <strong>latent nothing</strong> works well in most cases.</p><p><strong>Denoising strength</strong> controls how much changes it will make compared with the original image. Nothing will change when you set it to 0. You will get an unrelated inpainting when you set it to 1.</p><p>Make sure to <strong>generate a few images at a time</strong> so that you can choose the best ones. Set the <strong>seed</strong> to -1 so that every image is different.</p><p>Prompt: (Same as original)<br>Sampling steps: 20<br>Seed: -1<br>Image size: 704 x 512<br>Face restoration: Codeformer<br>Sampling method: Euler a<br>Model: Stable Diffusion v1.5 inpainting<br>Mask content: latent noise or latent nothing<br>Inpaint at full resolution: On<br>Denoising strength: 0.75</p><h3>Inpainting results</h3><p>Below are some of the inpainted images.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/0*84hkpwpD-IsdIRHs.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/0*DBgYTZrJF_PW07iD.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/0*rE_IMCqVzcfJPULB.png" /></figure><h3>One more round of inpainting</h3><p>I like the last one but there’s an extra hand under the newly inpainted arm. Following a similar step of uploading this image and creating a mask, the hand under the arm is removed with the second round of inpainting:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/0*xKo8RGa873EWxDv0.png" /><figcaption>Use inpainting to remove the extra hand under the arm.</figcaption></figure><p>And this is my final image.</p><p>A side-by-side comparison:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GHJptr0tuf6lvPQv.png" /><figcaption>Left: original. Right: inpainted 2 times.</figcaption></figure><p>As you have seen, inpainting can be an iterative process that you apply as many time you want to refine an image.</p><p>See this <a href="https://stable-diffusion-art.com/how-to-remove-a-person-with-ai-inpainting/">post</a> for another more extreme example of inpainting.</p><h3>Adding new objects</h3><p>Sometimes you want to add something new to the image.</p><p>Let’s try adding a hand fan to the picture.</p><p>First upload the image to inpainting canvas and create a mask around the chest and right arm.</p><p>Add prompt “holding a hand fan” to the beginning of the original prompt. The prompt for inpainting is</p><blockquote>(holding a hand fan: 1.2), [emma watson: amber heard: 0.5], (long hair:0.5), headLeaf, wearing stola, vast roman palace, large window, medieval renaissance palace, ((large room)), 4k, arstation, intricate, elegant, highly detailed</blockquote><p>Adding new objects to the original prompt ensures consistency in style. You can adjust the <a href="https://stable-diffusion-art.com/fine-tune-your-ai-images-with-these-simple-prompting-techniques/#Adjust_keyword_strength_using_weight">keyword weight</a> (1.2 above) to make the fan show.</p><p>Adjust <strong>denoising strength</strong> and <strong>CFG scale</strong> to fine-tune the inpainted images.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/0*8t0iANVAowSlUejc.png" /><figcaption>Adding a hand fan with inpainting.</figcaption></figure><h3>Explanation of inpainting parameters</h3><h4>Denoising strength</h4><p>Denoising strength controls how much respect the final image should pay to the original content. Setting it to 0 changes nothing. Setting to 1 you got an unrelated image.</p><p>Set to a low value if you want small change and high value if you want big change.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*eyGQq0acQwxIZuRs.png" /></figure><h4>Changing denoising strength. Set to low value if you want small change and high value if you want big change.<br><br>CFG scale</h4><p>Similar to usage in <a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#CFG_Scale">text-to-image</a>, <strong>Classifier Free Guidance scale</strong> is a parameter to control how much the model should respect your prompt.</p><p>1 — Mostly ignore your prompt.<br>3 — Be more creative.<br>7 — A good balance between following the prompt and freedom.<br>15 — Adhere more to prompt.<br>30 — Strictly follow the prompt.</p><h4>Masked content</h4><p>Masked content controls how the masked area are initialized.</p><ul><li><strong>Fill</strong>: Initialize with a highly blurred of the original image.</li><li><strong>Original</strong>: Unmodified.</li><li><strong>Latent noise</strong>: Masked area initialized with <strong>fill</strong> and random noise is added to the latent space.</li><li><strong>Latent zero</strong>: Like latent noise except no noise is added to the latent space.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*TbxUrceQLrhwLGPn.png" /><figcaption>Masked content.</figcaption></figure><h4>Inpaint at full resolution</h4><p>Inpaint the masked area at full resolution, then scale down to the original size. This helps to increase the quality of the inpainting.</p><p>When inpainting a small area, turning on full resolution is critical to the quality of the final image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/614/0*gcisJPknSPM_wzuD.png" /><figcaption>Inpaint at full resolution: Left: Off. Right: On.</figcaption></figure><p><em>This blog post is originally published on </em><a href="https://stable-diffusion-art.com/inpainting_basics/"><em>stable-diffusion-art.com</em></a></p><p>Visit to support this free blog:</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_prompt_generator?layout=profile">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li><li><a href="https://apps.apple.com/us/app/lucid-markdown-keyboard/id1616092954">Markdown Keyboard for iPhone and iPad</a></li><li><a href="https://apps.apple.com/us/app/notion-keyboard-notionkey/id6443495034">Notion Keyboard — Notionkey for iPhone and iPad</a></li><li><a href="https://apps.apple.com/us/app/auto-fonts-keyboard/id1636807987">Auto Font Keyboard for iPhone &amp; iPad</a></li></ul><p>Follow me on <a href="https://twitter.com/andrewwongai">Twitter</a> &amp; <a href="https://www.instagram.com/andrewwongai/">instagram</a> to receive content like this.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=26445f31248a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Absolute beginner’s guide for Stable Diffusion AI image]]></title>
            <link>https://medium.com/@andrewwongai/absolute-beginners-guide-for-stable-diffusion-ai-image-31f97d2eb070?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/31f97d2eb070</guid>
            <category><![CDATA[stable-diffusion-art]]></category>
            <category><![CDATA[ai-art-generator]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Sat, 12 Nov 2022 12:16:27 GMT</pubDate>
            <atom:updated>2022-11-12T12:23:59.027Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Absolute beginner&#39;s guide for Stable Diffusion AI image" src="https://cdn-images-1.medium.com/max/704/0*MFbJU1BMvhJMiDG_.png" /></figure><h3>About this Beginner’s guide</h3><p>This beginner’s guide is for absolute newcomers who have zero experience with Stable Diffusion or other AI image generators.</p><p>I will cover:</p><ul><li>What Stable diffusion can do for you.</li><li>The quickest way to start generating images.</li><li>Basics of text prompts and parameters.</li><li>How to get better at engineering prompts.</li><li>Common ways to fix defects.</li></ul><h3>What is Stable Diffusion?</h3><p>Stable Diffusion is an AI model that generate image from text input. Let’s say if you want to generate images of a gingerbread house, you can put in the prompt:</p><blockquote>gingerbread house, diorama, in focus, white background, toast , crunch cereal</blockquote><p>The AI model will generates images that matches the prompt:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*X4ldmkqRpjY8gjAH.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*xl2MshTm3mKsH-yx.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*qY95xXg_XuBW9osj.png" /></figure><p>The advantage of Stable Diffusion is that</p><ul><li>It is open-source: Many enthusiasts have created free and powerful tools.</li><li>It is designed for low computational resources: It’s free or cheap to run.</li></ul><h3>How to start generating images?</h3><h4>Online generator</h4><p>For absolute beginners, I recommend using a <a href="https://stable-diffusion-art.com/free-ai-image-generator-sites/">free online generator</a>. Go to one of the sites in the list, putting in the example prompt above and you are in business!</p><p>The fastest way to generate high quality images is to reuse existing prompts. Head to the <a href="https://stable-diffusion-art.com/prompts/">prompt collection</a> section, pick an image you like and steal the prompt!</p><p>Alternatively, use image collection sites like <a href="https://lexica.art/">Lexica</a>. Pick an image you like and use the prompt.</p><p>Treat the prompt as a starting point. Modify to suit your needs.</p><p>Advanced GUI</p><p>The downside of online generators is that the functionality is pretty limited.</p><p>If you outgrew them, you should move on to using a more advanced GUI (Graphical User Interface). I use <a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui">AUTOMATIC1111</a> which is a powerful and popular choice. See <a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">my guide</a> for setting up in Google Colab cloud server.</p><p>With advanced GUI, you can explore fine-tuning with your images by using more advanced techniques like <a href="https://stable-diffusion-art.com/fine-tune-your-ai-images-with-these-simple-prompting-techniques/#Blend_two_keywords">keyword blending</a> and <a href="https://stable-diffusion-art.com/how-to-remove-a-person-with-ai-inpainting/">inpainting</a>.</p><h4>How to build good prompts?</h4><p>Two advices: (1) Be as detailed and specific, and (2) use powerful keywords.</p><p>Be detailed and specific in your prompt</p><p>The first rule is be as specific. Stable Diffusion cannot read your mind. You need to describe your image in as much detail as possible.</p><p>For example, if you want to generate a picture of a woman sitting on a chair, instead of using the prompt</p><blockquote>A woman sitting on a chair</blockquote><p>(Prompt #1)</p><p>You should use something like:</p><blockquote>a young gorgeous woman with long blonde hair sitting on a 19th-century antique chair, beautiful eyes, symmetric face, dedicated modern clothing, photo</blockquote><p>(Prompt #2)</p><p>See the drastic difference in image quality between the two prompts:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iTWYq06X1QuGeU46.png" /><figcaption>Example images from simplistic prompt #1 (left) and detailed prompt #2 (right).</figcaption></figure><p>So work on your prompts before writing off Stable Diffusion!</p><p>Use powerful keywords in your prompt</p><p>Some keywords are just more powerful than others. Examples are</p><ul><li>Celebrity name (e.g. Emma Watson)</li><li>Artist name (e.g. van Gogh)</li><li>Genre (e.g. illustration, painting, photograph)</li></ul><p>Using them carefully can steer the image to the direction you want.</p><p>You can learn more about prompt building and example keywords in <a href="https://stable-diffusion-art.com/how-to-come-up-with-good-prompts-for-ai-image-generation/">basics of building prompt</a>.</p><h4>What do the parameters mean and should I change them?</h4><p>Most<a href="https://stable-diffusion-art.com/free-ai-image-generator-sites/"> online generators</a> allow you to change a limited set of parameters. Below are some important ones:</p><ul><li><a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#Image_size"><strong>Image size</strong></a>: The size of output image. The standard size is 512x512 pixels. Changing it to portrait or landscape size can have big impact on the image. For example, use <a href="https://stable-diffusion-art.com/common-problems-in-ai-images-and-how-to-fix-them/#Not_showing_full_body_problems">portrait size </a>if you want to generate a full-body image.</li><li><a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#Sampling_steps"><strong>Sampling step</strong></a><strong>s</strong>: Use at least 20 steps. Increase if you see blurry image.</li><li><a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#CFG_Scale"><strong>CFG scale</strong></a>: Typical value is 7. Increase if you want the image to follow the prompt more.</li><li><a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#Seed"><strong>Seed value</strong></a>: -1 if you want a random image. Specify a value if you want the same image.</li></ul><p>See <a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/">recommendations of other settings</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/0*j49U83pUctkMuy9A.png" /></figure><h3>How many images should I generate?</h3><p>You should always generate multiple images when testing a prompt.</p><p>I generate 4 images at a time when making big changes to the prompt, so that I can speed up the search. I would generate 8 at a time when making small changes to increase the chance I would see something usable.</p><p>Some prompts only work half of the time or less. So don’t write off a prompt based on one image.</p><h3>Common ways to fix defects in images</h3><p>When you see stunning images shared in social media, there’s a good chance that they have undergone series of post-processing steps.</p><p>Face restoration</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*IDZLL17enFVhX2FF.png" /><figcaption>Left: Original images. Right: After face restoration.</figcaption></figure><p>It’s well known in the AI artist community that Stable Diffusion is not good at generating faces. Very often the faces generated have artifacts.</p><p>We often use image AI models that specifically trained for restoring faces, for example <a href="https://github.com/sczhou/CodeFormer">CodeFormer</a> which AUTOMATIC1111 GUI has built in support. See <a href="https://stable-diffusion-art.com/common-problems-in-ai-images-and-how-to-fix-them/#Garbled_faces_and_eyes_problems">how to turn it on</a>.</p><p>Fixing small artifacts with inpainting</p><p>It is not easy to get an image you want in one shot. A better approach is to aim at generating an image with good composition and repair the defects.</p><p>Below is an example of an image before and after inpainting. Using the original prompt for inpainting would work 90% of the time. Make sure to turn on high resolution fix.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*bGtJvW-wANM-RWOr.png" /><figcaption>Left: Original image with defects. Right: The face and arm are fixed by inpainting.</figcaption></figure><p>Read more about ways to <a href="https://stable-diffusion-art.com/common-problems-in-ai-images-and-how-to-fix-them/">fix common issues</a>.</p><p>That’s it for now! Hope you enjoy this article.</p><p>This article is original in <a href="https://stable-diffusion-art.com/beginners-guide/">stable-diffusion-art.com</a>.</p><p>Support this free blog:</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_prompt_generator?layout=profile">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li><li><a href="https://apps.apple.com/us/app/lucid-markdown-keyboard/id1616092954">Markdown Keyboard for iPhone and iPad</a></li><li><a href="https://apps.apple.com/us/app/notion-keyboard-notionkey/id6443495034">Notion Keyboard — Notionkey for iPhone and iPad</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=31f97d2eb070" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[4 methods to generate fashion ideas with image AI]]></title>
            <link>https://medium.com/@andrewwongai/4-methods-to-generate-fashion-ideas-with-image-ai-e3c4a362bde1?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/e3c4a362bde1</guid>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[inpainting]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[ai-art-generator]]></category>
            <category><![CDATA[stable-diffusion-art]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Tue, 08 Nov 2022 06:14:04 GMT</pubDate>
            <atom:updated>2022-11-08T13:31:19.755Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="4 methods to generate fashion ideas with image AI" src="https://cdn-images-1.medium.com/max/600/0*Qi9xcNbrJdTNVkFl.png" /></figure><p>Stable Diffusion AI distills millions of images in the model. In layman’s term, it has seen a lot. If creativity is new arrangement of knowledge, it surely has potential to create.</p><p>In this post, we will use the image AI to generate new fashion clothings. The AI will show us something we have never seen, and we can use them as ideas for our own projects.</p><h4>Software</h4><p>We will use Stable Diffusion AI and AUTOMATIC1111 GUI. See my <a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">quick start guide</a> for setting up in Google’s cloud server.</p><h4>Select a starting image</h4><p>You will first need to get a fashion model. The best is to get a full-body image with a simple monotonic background. Below is what I am going to use in this tutorial.</p><figure><img alt="Starting fashion model." src="https://cdn-images-1.medium.com/max/560/0*bDqIlOKB3oU6bQap.png" /><figcaption>Original image</figcaption></figure><h4>Create inpaint mask</h4><p>In the Stable Diffusion GUI, go to <strong>img2img</strong> tab and select the <strong>inpaint</strong> tab.</p><p>Upload the starting image by dragging and dropping it to the inpaint image box.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*IU_KaX2O1e7An-UN.png" /></figure><p>Then create a mask covering the whole body like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/264/0*OU89Uo-FjsshkJDM.png" /></figure><p>Use <em>crop and resize</em>, and set the <em>image size</em> to aspect ratio similar to the original.</p><p>In the methods below, you may want to explore changing <a href="https://stable-diffusion-art.com/know-these-important-parameters-for-stunning-ai-images/#CFG_Scale">CFG</a> to strike a balance between freedom and following the prompt.</p><h4>Method 1: Describing the apparel items</h4><p>Use terms describing what the model is wearing is an effective way of generating a specific clothing.</p><p>For example, images below are using a single-word prompt “sweater” for inpainting. (Euler a, 35 steps, denoising strength 0.9, CFG 7).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*XHIvo6mZuhtOPZHB.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*toChoAx8iPkLoWsk.png" /></figure><p>But you don’t need to stop there. You can throw a kitchen sink at Stable Diffusion and let it “figure out”. Below is a prompt that includes multiple fashion items.</p><p>Prompt:</p><blockquote>colorful, wedding dress, Sweater, T-shirt, Sheath dress, Dress, Gym clothes, Bra, Tank top, Shorts, Hoodie, Jeans, Long coat, Uniform, Coat, Dress pants, Swimsuit, Long-sleeve, top, Skirt, Thong, Hat, Necklace, Ring, blue, yellow, green, black, Earrings,Perfume, Mittens, Sunglasses, Handbag, Bracelet, Purse, Watch, Umbrella, Scarf, Slip, High-heeled shoes, Flip-flops, Boots, Stockings, Jacket</blockquote><p>The results are definitely interesting:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*W2S9Bjy0vdZATQR2.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*DuNnlbqptEOVKzHM.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*lXAIQgj72D1Rv3ng.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*QJm6mrDUYrqMNjrD.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*tHxzaic52OX_b6S_.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*L_3O7f-Z1-2C2p7K.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*4UPTw2Q3Z8a0fFPY.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*rJ3qTQiBiVqRG3uH.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*5sj4WuI9eVz7XVYz.png" /></figure><p>When using diverse apparel terms and relatively low CFG (7), we allow Stable Diffusion AI to pick and choose the combinations that makes most sense.</p><h4>Method 2: Apparel types</h4><p>We can use apparel categories as prompt. That avoids being too specific and can create an element of surprise.</p><p>Below are generated with prompt “Sun dress”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*fMuibfP9yS_7LMtn.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*KkJ394ENeCmLXu66.png" /></figure><p>Here’s a list of other apparel types, just to give you some idea!</p><p>- Culotte dress<br>- Babydoll dress<br>- Wrap around dress<br>- Kimono dress<br>- Bubble dress<br>- Tutu dress<br>- Polo dress<br>- Shirtdress<br>- Sundress<br>- Sari<br>- Sack dress<br>- Tunic dress<br>- Sheath dress<br>- Shift dress<br>- Trumpet dress<br>- Smoked/layered dress</p><h4>Method 3: Emotional words</h4><p>Somethings it is better to say how great it is without saying what it is. Let Stable Diffusion figure out the fashion.</p><p>Prompt:</p><blockquote>elegant, stylish, stunning</blockquote><p>(denoising 1, cfg 15)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*2gbJ9vnk7cV2jFDr.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*n8ca4QTLJCMS4IQl.png" /></figure><h4>Method 4: Occupation</h4><p>Let’s include uniforms in our fashion brainstorming. The trick is to think about what kind of people wear uniform for their work.</p><p>Prompt:</p><blockquote>Doctor</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*p-qb4ARnJkotKX-V.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*HNV1paFtkZqys4Co.png" /></figure><p>Prompt:</p><blockquote>Flight attendant</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*tuvFKgRwJK2FsgqS.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*_qLjlPV_8miGW9Oy.png" /></figure><p>Some other ideas are</p><p>- Firefigther<br>- Pumber<br>- Maid<br>- Chef<br>- Pilot<br>- Lifeguard</p><h4>Summary</h4><p>We have covered using inpainting to create different fashion styles using the free and open-source Stable Diffusion AI. Describing the fashion items directly is a powerful method but not the only way.</p><p>Sometimes it makes sense to be less specific and let AI explore the space.</p><p>Hope you will find this tutorial useful.</p><p>Follow me on <a href="https://twitter.com/andrewwongai">Twitter</a> to receive content like this.</p><p>Subscribe to my <a href="https://www.sagiodev.com/newsletters/">newsletter</a> to receive weekly updates.</p><p>Support this free blog by using my products</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_prompt_generator?layout=profile">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li><li><a href="https://apps.apple.com/us/app/lucid-markdown-keyboard/id1616092954">Markdown Keyboard for iPhone and iPad</a></li><li><a href="https://apps.apple.com/us/app/notion-keyboard-notionkey/id6443495034">Notion Keyboard — Notionkey for iPhone and iPad</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e3c4a362bde1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Make an animated GIF with Stable Diffusion (step-by-step)]]></title>
            <link>https://medium.com/@andrewwongai/step-by-step-guide-to-make-an-animated-gif-with-stable-diffusion-633b801ab44c?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/633b801ab44c</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[tutorial]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[inpainting]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Mon, 07 Nov 2022 00:26:49 GMT</pubDate>
            <atom:updated>2022-11-07T13:57:29.893Z</atom:updated>
            <content:encoded><![CDATA[<p>Stable Diffusion AI is a great way to synthesize images from text. Do you can spice up your artworks with animation?</p><p>After this tutorial, you will be able to recreate this final animated GIF image:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*wo2RfLNlFueNNDsNs6jJmQ.gif" /><figcaption>Animated waterfall and moving clouds added.</figcaption></figure><p>In this step-by-step guide, we will add an animated waterfall and moving clouds to the image of a fantasy steam punk castle shown below. The prompt for generating this image can be found in <a href="https://stable-diffusion-art.com/steam-punk-castle/">this page</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*NOe0Y3CrzIukpbecLrhe1w.png" /><figcaption>Original image of a fantasy steam punk castle.</figcaption></figure><h3>Stable Diffusion GUI</h3><p>We will need to use a more advanced Stable diffusion GUI for this tutorial. Most online generators won’t cut it. We will use <a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui">this</a> GUI. See my <a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">quick start guide</a> for setting up in Google’s cloud server.</p><h3>Starting image</h3><p>The prompt and parameters necessary to generate the original image can be found in <a href="https://stable-diffusion-art.com/steam-punk-castle/">this page</a>. Alternatively, you can save the following image to your computer for this tutorial.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*NOe0Y3CrzIukpbecLrhe1w.png" /><figcaption>Starting image for this tutorial.</figcaption></figure><h3>Step 1: Add a waterfall</h3><p>We will use <a href="https://stable-diffusion-art.com/how-to-remove-a-person-with-ai-inpainting/">inpainting</a> to add the waterfall.</p><p>In the Stable Diffusion GUI, go to <strong>img2img</strong> tab and select the <strong>inpaint</strong> tab.</p><p>Upload the starting image by dragging and dropping it to the inpaint image box.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pP8cz1xEjUIiCCUjWxzFtw.png" /></figure><p>Use the paintbrush tool to create a mask like below. The masked area is where the waterfall will be located.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/486/1*Kx3T6U7Q7EQb7P0cGvkS-A.png" /><figcaption>Inpaint mask for the waterfall.</figcaption></figure><p>Put in the following settings:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NKdWIKKcZRkQlwqHvX9Iog.png" /></figure><p>Hit <strong>Generate</strong>. You should get 8 different images with a waterfall added.</p><p>Increase CFG value if you see an unwanted object. Decrease it if you prefer to have more variations.</p><p>Hit generate again until you see an image you like. Save that image to your computer.</p><p>I picked the image below for the castle image with waterfall added.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*2UuMqiMI7oEAOh6jezg21w.png" /><figcaption>Waterfall added by inpainting.</figcaption></figure><p>This image will be our <strong>base image</strong> for the animated GIF. In the next step, we will create small variations of this image and use them as the frames for the animated GIF.</p><h3>Step 2: Animate the water fall and the clouds</h3><p>An animated GIF is nothing but a series of images displayed consecutively. The variations between images should be subtle but noticeable in order to create a perception of animation.</p><p>To do so, we first drag and drop the castle image with the waterfall to the inpaint box.</p><p>Create the inpaint mask like below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*5uM9h4QX6P2oyBm7SvZq0w.png" /><figcaption>Inpaint mask for animating the cloud and the waterfall.</figcaption></figure><p>Use the following parameters for <strong>inpainting</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KIloRYKtQ2Y_aV1XIq8YhQ.png" /></figure><p>Recall we don’t want to change the image too much so using a low denoising strength like 0.2 is very important. (Recall that denoising strength of 0 changes nothing and at 1 your original image in the masked area will be completely ignored.)</p><p>Below are a few images I generated. If you pay close attention, you should see they have small differences.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*ygesRt4TRVmORuuKXYFtpA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*uj0QFFMByCkYL7WGbaLGfg.png" /></figure><p>You will want to cherry-pick the images you want to use for the animated GIF.</p><p>You will need 5 to 10 images. Save them to your computer.</p><p>Increase denoising strength if you want more variations. Decrease if you want less.</p><h3>Step 3: Create the animated GIF</h3><p>Now we will make the animated GIF.</p><p>You can use any tool of your choice but in this tutorial we will use the free site <a href="https://ezgif.com/maker">ezgif.com</a>. Upload the images of your choice by using the “Choose Files” button and then clicking the “Upload and make a GIF!” button.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WmvRUzY6UOONkpi-vIDeag.png" /></figure><p>In the next screen, check the “crossfade frames” under the “Effects” section and click “Make a GIF!”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TfPv3nBIuNTkhe37Pf0bLQ.png" /></figure><p>Finally, save the animated GIF. The animated GIF should look like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*wo2RfLNlFueNNDsNs6jJmQ.gif" /><figcaption>Animated waterfall and moving clouds added.</figcaption></figure><p>For comparison, this is the animated GIF created with denoising strength 0.1.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*C5t8cgdBQuijN2c4r33g8w.gif" /><figcaption>Animated GIF created with denoising strength 0.1. The variation is more subtle.</figcaption></figure><h3>Tweaks for your own artwork</h3><p>Note that the settings in this article are specific for the image we used. You may need to play with the following parameters when you create your own artworks.</p><ul><li><strong>Prompts</strong> for inpainting: Change to what you want to add in place of the mask.</li><li><strong>CFG</strong>: Increase if you see irrelevant objects or too much variations.</li><li><strong>Denoising strength</strong>: Increase if you want more changes in your animated GIF.</li><li><strong>GIF delay time</strong> and <strong>crossfade parameters</strong>: Adjust to smooth out the animations.</li></ul><p>In this tutorial, we detailed how to use inpainting to create images for making an animated GIF. I hope this serves as a good starting point for you to create your own artworks.</p><p>Don’t hesitate to drop me a line if you have any questions!</p><p><em>Originally published at </em><a href="https://stable-diffusion-art.com/step-by-step-guide-to-make-an-animated-gif-with-stable-diffusion/"><em>https://stable-diffusion-art.com</em></a><em> on November 7, 2022.</em></p><p>Follow me on <a href="https://twitter.com/andrewwongai">Twitter</a> to receive content like this.</p><p>Subscribe to my <a href="https://www.sagiodev.com/newsletters/">newsletter</a> to receive weekly updates.</p><p>Support this free blog by using my products</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_prompt_generator?layout=profile">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=633b801ab44c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to fix these common issues in AI images]]></title>
            <link>https://medium.com/@andrewwongai/how-to-fix-these-common-issues-in-ai-images-9dadc0688cf3?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/9dadc0688cf3</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[imageai]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[ai-image-generator]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Tue, 01 Nov 2022 04:54:17 GMT</pubDate>
            <atom:updated>2022-11-29T13:13:09.953Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*KH51GQNjwKxVGrm8.png" /></figure><p>Seeing some funny issues in AI images? I will teach you how to fix some of the common ones.</p><p><em>(This article was first published in </em><a href="https://stable-diffusion-art.com/"><em>https://stable-diffusion-art.com/</em></a><em>)</em></p><h3>Software</h3><p>We will use <a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui">this</a> Stable Diffusion GUI for this tutorial. See my <a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">quick start guide</a> for setting up in Google’s cloud server.</p><h3>Two heads</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*dD0Ii9k4x6Uh_ZU6.png" /></figure><p>It’s not unusual to see images with two heads connecting together in Stable Diffusion. It is usually caused by using portrait image size. To avoid it, use image size with 1:1 aspect ratio, for example 512x512.</p><p>Three ways to tackle the two-head issue:</p><ol><li>Generate multiple images and discard those with two heads.</li><li>Reduce aspect ratio. The closer to 1:1 ratio, the less likely you would encounter the infamous two-head.</li><li>Generate a full-body image. (See next section for techniques.)</li></ol><h3>Not showing full body</h3><p>A common suggestion to get a full-body image is to use the keyword “full body portrait”. But more often than not it just doesn’t work. Below are a few examples with “full body portrait” keyword added. You can see the keywords are completely ignored.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*vTj9De0FbxaEbEtP.png" /></figure><p>A more consistent approach is to add keywords of what you want to <em>see</em> in the picture. For example</p><ul><li>standing</li><li>long dress</li><li>legs</li><li>shoes</li></ul><p>Belows are generated with adding keywords <em>standing</em> and <em>long dress</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*msaRrB0gRknouKev.png" /></figure><p>Now it is showing more of the body but we can do better. The last trick I have is to change image size to a portrait size on top of using these keywords. Below are some examples. That’s all I know to tease full-body images out of the model!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iib7mejnsGqOzHPZ.png" /></figure><h3>Garbled faces and eyes</h3><p>As detailed in a previous article, it is fairly common to see distorted faces. In fact it happens to other body parts, just that our brain is particularly sensitive to any asymmetry in faces.</p><p>To fix faces and eyes, turn on face restoration if your GUI supports it. See my <a href="https://medium.com/@andrewwongai/control-your-ai-images-with-these-must-know-parameters-e5cfbfba192f">previous post</a> for detail. If you are using a <a href="https://medium.com/@andrewwongai/free-ai-image-generation-sites-9344c8df0aeb">free site</a> with no built-in support for face restoration, you use the following sites for post-processing.</p><ul><li><a href="https://huggingface.co/spaces/Xintao/GFPGAN">GFPGAN</a></li><li><a href="https://huggingface.co/spaces/sczhou/CodeFormer">CodeFormer</a></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sDKxtz8SiNkql7Fq.png" /></figure><h3>Messed-up fingers</h3><p>It’s well-known in stable diffusion community that the AI has problem getting fingers right. It is not uncommon to see distorted, too many or too few fingers.</p><p>The first fix is to include keywords that describe hands and fingers like “beautiful hands” and “detailed fingers”. That tends to prime the AI to include hands with good details.</p><p>The second fix is to use <a href="https://medium.com/@andrewwongai/how-to-remove-unwanted-objects-in-photos-with-ai-inpainting-72ab5c8f7f1f">inpainting</a>. Create a mask in the problematic area. Use inpainting to generate multiple images and choose the one you like. I found it sufficient to reuse the same prompt as the original image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cGhI8D_Lg_4hXt7L.png" /></figure><p>Thats all for today. Hope you find these techniques useful!</p><p>Follow me on <a href="https://twitter.com/andrewwongai">Twitter</a> to receive content like this.</p><p>Subscribe to my <a href="https://www.sagiodev.com/newsletters/">newsletter</a> to receive weekly updates.</p><p>Support this free blog by using my products</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_prompt_generator?layout=profile">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li><li><a href="https://apps.apple.com/us/app/lucid-markdown-keyboard/id1616092954">Markdown Keyboard for iPhone and iPad</a></li><li><a href="https://apps.apple.com/us/app/notion-keyboard-notionkey/id6443495034">Notion Keyboard — Notionkey for iPhone and iPad</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9dadc0688cf3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Control your AI images with these must-know parameters]]></title>
            <link>https://medium.com/@andrewwongai/control-your-ai-images-with-these-must-know-parameters-e5cfbfba192f?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/e5cfbfba192f</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[digital-images]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[ai-image-generator]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Fri, 28 Oct 2022 18:27:10 GMT</pubDate>
            <atom:updated>2022-10-28T18:27:10.850Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/876/0*-DJQt4OoTUKuROjJ.png" /></figure><p>So you have generated a few AI images. They looked great but not quite what you want? You can use some customization. Here’s a primer for the basic generation parameters.</p><h3>Software</h3><p>We will focus on Stable Diffusion AI. While some parameters mentioned in this article are available in <a href="https://medium.com/@andrewwongai/free-ai-image-generation-sites-9344c8df0aeb">free online AI generators</a>, all of them are available in <a href="https://github.com/AUTOMATIC1111/stable-diffusion-webui">this</a> popular Stable Diffusion GUI (AUTOMATIC1111). See my <a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">quick start guide</a> for setting up in Google’s cloud server.</p><h3>CFG Scale</h3><p>Classifier Free Guidance scale is a parameter to control how much the model should respect your prompt.</p><p>1 — Mostly ignore your prompt.<br>3 — Be more creative.<br>7 — A good balance between following the prompt and freedom.<br>15 — Adhere more to prompt.<br>30 — Strictly follow the prompt.</p><p>Below are a few examples of increasing the CFG scale with the same random seed. In general, you should stay away from the two extremes — 1 and 30.</p><p>Recommendation: Starts with 7. Increase if you want it to follow your prompt more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*FjYdJOFPrUX-KSuG.png" /><figcaption>Changing CFG scale.</figcaption></figure><h3>Sampling steps</h3><p>Quality improves as the sampling step increases. Typically 20 steps with <em>Euler a</em> sampler is enough to reach a high quality, sharp image. Although the image will still change subtly when stepping through to higher values, it will become different but not necessarily higher quality.</p><p>Recommendation: 20 steps. Adjust to higher if you suspect quality is low.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*TU1AasW5rhwXz9u9.png" /><figcaption>Changing number of sampling steps.</figcaption></figure><h3>Sampling methods</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*75XVSbJ6YOpynJW6.png" /><figcaption>Sampling methods available in AUTOMATIC1111.</figcaption></figure><p>There’s a variety of sampling methods you can choose, depending on what GUI you are using. They are simply different methods for solving diffusion equations. They are supposed to give the same result but could be slightly different due numerical bias. But since there’s no right answer here — the only criteria is the image looks good, accuracy of the method should not be your concern.</p><p>There are discussions in online community claiming that certain sampling methods tend to yield particular styles. This is without theoretical merit. My suggestion is to leave it as the default. You would only want to change it if you want to speed things up.</p><p>Recommendation: leave it as default.</p><h3>Seed</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/0*ccLEw_WxthduBpsw.png" /><figcaption>Random seed input box.</figcaption></figure><p>The random seed determines the initialize noise pattern and hence the final image.</p><p>Setting it to -1 means using a random one every time. It is useful when you want to generate new images. On the other hand, fixing it would result in the same images in each new generation.</p><p>How to find the seed used for an image if you use random seed? In the dialog box, you should see something like:</p><blockquote>Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 4239744034, Size: 512x512, Model hash: 7460a6fa</blockquote><p>Simply copy this seed value to seed input box. If you generate more than one image at a time, the seed value of the second image is this number incremented by 1, and so on.</p><p>Recommendation: Set to -1 to explore. Fix to a value for fine-tuning.</p><h3>Image size</h3><p>The size of output image. Since Stable Diffusion is trained with 512x512 images, setting it to portrait or landscape sizes can create unexpected issues. Leave it as square whenever possible.</p><p>Recommendation: Set image size as 512x512.</p><h3>Batch size</h3><p>Batch size is the number of images generated each time. Since the final images are very dependent on the random seed, it is always a good idea to generate a few images at a time. This way, you can get a good sense of what the current prompt can do.</p><p>Recommendation: Set batch size to 4 or 8.</p><h3>Restore faces</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*naz0F8dGGWqzvAit.png" /><figcaption>Restore faces checkbox is shown on the left.</figcaption></figure><p>A dirty little secret of Stable Diffusion is that it often has issues with faces and eyes. Restore faces is a post-processing method applied to images using AI trained specifically to correct faces.</p><p>To turn it on, check the box next to Restore faces. Go to Settings tab, under Face restoration model, select CodeFormer.</p><p>Below are two examples. Without face restoration is on the left. With face restoration is on the right.</p><p>Recommendation: Turn restore faces on when you generate images with faces.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*STiLwrpFeAKkUbFs.png" /><figcaption>Left: Original images. Right: After face restoration.</figcaption></figure><h3>Summary</h3><p>In this article, we have covered the basic parameters for Stable Diffusion AI. Hope you find it useful!</p><p>Follow me on <a href="https://twitter.com/andrewwongai">Twitter</a> to receive content like this.</p><p>Subscribe to my <a href="https://www.sagiodev.com/newsletters/">newsletter</a> to receive weekly updates.</p><p>Support this free blog by using my products</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/prompt_builder_for_stable_diffusion_premium">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://apps.apple.com/us/app/lucid-markdown-keyboard/id1616092954">Markdown Keyboard for iPhone and iPad</a></li><li><a href="https://apps.apple.com/us/app/notion-keyboard-notionkey/id6443495034">Notion Keyboard — Notionkey for iPhone and iPad</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e5cfbfba192f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[3 things I did before app launch and what I wish I have done]]></title>
            <link>https://medium.com/@andrewwongai/3-things-i-did-before-app-launch-and-what-i-wish-i-have-done-147aca96caf4?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/147aca96caf4</guid>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[prelaunch]]></category>
            <category><![CDATA[ios-app-development]]></category>
            <category><![CDATA[app-launch]]></category>
            <category><![CDATA[app-marketing-tips]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Fri, 21 Oct 2022 11:26:18 GMT</pubDate>
            <atom:updated>2022-10-21T11:36:17.705Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/200/0*Xp3tq-TGMBXAOQOu.png" /></figure><p>Earlier this month I launched <a href="https://apps.apple.com/us/app/notion-keyboard-notionkey/id6443495034">Notionkey</a>, an iOS custom keyboard designed to make writing easier in Notion. This is the third iOS app I launched. This time I wanted to do the launch right so I spent some marketing effort in pre-launch.</p><p>This is a write-up of what I did, the outcome, and what I wish I have done.</p><h3>Product Hunt</h3><p>I decided to launch on <a href="https://www.producthunt.com/products/notionkey">Product Hunt</a> because I heard good words about it on Twitter. For people who are not familiar with it, it’s a website devoted to new digital products, be it a website, a desktop program, or in my case an iPhone app.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*ygEEuA97D7qG_xhx.png" /></figure><p>You cannot actually <em>launch</em> a product on Product Hunt per se, my app is still only available on Apple’s App Store. This is more for marketing purpose so that people can know about your app when it is launched. On launch day, you make your product page available for vote. The products launched on the same day are then ranked by upvotes. The products ranked high got the benefit of getting lots of traffics.</p><p>However, I immediate found that I was already late to the game because the number of upvotes are quite related to the number of followers you have. True, it does not matter if you have a killer product that the community cares about. But followers will very likely upvote your product.</p><p>Because of the voting mechanism, people will have more time to vote when you schedule your launch. So make sure you don’t <em>launch</em> right away even if your app is already available.</p><p>The app did ok, gathered <a href="https://www.producthunt.com/products/notionkey">40 something upvotes</a> on the launch day. Enough to place the app within top 20.</p><h3>Take home for Product Hunt</h3><ul><li>Should have been active in Product Hunt 3 months before launch.</li><li>Overall still quite useful for newcomers</li></ul><h3>Twitter</h3><p>About a month before launch, I decided to start building my <a href="https://twitter.com/andrewwongai">twitter account</a> to generate organic traffic. At launch I may have like 50 followers…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*i5hnMMVkSlySLjn3.png" /></figure><p>During development, I tweeted a demo video to the Notion community and asked who want it free at launch, and separately who wanted to be beta testers. Twitter works in a way that if your post generates good reaction, the tweet would be amplified exponentially. Those two tweets received a lot more attention than my number of followers deserved. I gained some followers and recruited about 10 beta testers. But I think the result would be greater if my followers are in the thousands.</p><h3>Take home for Twitter</h3><ul><li>Giveaway tweets are effective</li><li>Recruit beta tester in niche community</li><li>Should have built Twitter profile early</li></ul><h3>Reddit</h3><p>Out of the three social media, my effort in Reddit is the most successful one. In fact, I used Reddit to validate my app idea.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*vkDFYbzVRsKNS3MH.png" /></figure><p>In the very beginning, I conducted a <a href="https://www.reddit.com/r/Notion/comments/xbwa9k/do_you_found_it_hard_to_format_text_in_notions/">poll</a> in Notion subreddit and validated the idea that writing in Notion mobile app is really a pain. 90% of respondents said it was hard to reach the formatting buttons. I knew I was on to something and started building the app.</p><p>Mid-way in development, I posted a demo video of a working prototype and received 200+ likes. On launch day, I posted another update and got 400+ likes and a lot of compliments.</p><p>Two factors contributed to the success. First, Reddit is organized in special interest groups which allowed me to reach a niche audience. My post got a lot of likes because I am solving a problem the Notion community cares about. Second, unlike Twitter, content, not connection, matters. It’s a more merit-based platform. Your post will go viral if it is something important to the community. In contrast, it’s hard for a tweet from a small account to go viral in Twitter, unless it is retweeted by a big account.</p><p>One drawback of reddit is, unlike Twitter, you cannot post anything you want. You have to be careful and read the rules of the subreddit. Although my launch day post went viral and got a lot of likes, and I thought it was ok to post, the mod eventually determined I violated the self-promotion rule. The post got deleted and I got banned for 2 weeks. To the credit of the mod, they only did that a week after the launch day when the app has already reaped all the traffics. If I were to do it again, I would make the app free for the first two weeks so the post stays within the rules.</p><h3>Take home for reddit</h3><ul><li>Use poll to validate your idea</li><li>Get feedback by posting demos</li><li>Read the rules of the subreddit</li></ul><h3>Summary</h3><p>So what’s the final outcome? A classic launch-day spike and more importantly users who find the app useful and stick around. It’s a great feeling to know I have made something useful to other people, be it an app, a Notion template, a blog post or a tweet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*slXZV_zEWgV8BLks.png" /></figure><p>To sum up, Reddit is the most successful approach for me. In part because it’s more democratic than other platforms. It doesn’t need any followers for your post to go viral. What counts is the content. On the other hand, if I were to have more followers on the other two platforms, they could also be equally if not more successful.</p><p><em>All artworks in this post are made with Stable Diffusion.</em></p><p>Follow me on <a href="https://twitter.com/andrewwongai">Twitter</a> to receive content like this.</p><p>Subscribe to my <a href="https://www.sagiodev.com/newsletters/">newsletter</a> to receive weekly updates and exclusive content.</p><p>Support me by using my products</p><ul><li><a href="https://apps.apple.com/us/app/notion-keyboard-notionkey/id6443495034">Notion Keyboard — Notionkey for iPhone and iPad</a></li><li><a href="https://apps.apple.com/us/app/lucid-markdown-keyboard/id1616092954">Markdown Keyboard for iPhone and iPad</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/prompt_builder_for_stable_diffusion_premium">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage?layout=profile">Stable Diffusion AI artwork shuffling collage</a></li></ul><p>You may also like:</p><p><a href="https://medium.com/@andrewwongai/read-this-before-working-on-your-first-ios-app-e293456e67fc">Read this before working on your first iOS app</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=147aca96caf4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[11 free AI image generation sites you should know]]></title>
            <link>https://medium.com/@andrewwongai/free-ai-image-generation-sites-9344c8df0aeb?source=rss-f91d8ad9e8a2------2</link>
            <guid isPermaLink="false">https://medium.com/p/9344c8df0aeb</guid>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[ai-art-generator]]></category>
            <category><![CDATA[prompt]]></category>
            <category><![CDATA[free]]></category>
            <dc:creator><![CDATA[Andrew]]></dc:creator>
            <pubDate>Wed, 19 Oct 2022 19:35:05 GMT</pubDate>
            <atom:updated>2023-02-20T23:56:18.991Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*8zbQcScwne9HbkGFdTwUrw.png" /></figure><p>Upate Nov 10, 2022: This list is now maintained here:</p><p><a href="https://stable-diffusion-art.com/free-ai-image-generator-sites/">https://stable-diffusion-art.com/free-ai-image-generator-sites/</a></p><p>Cost of generating AI images has plummeted since the release of Stable Diffusion. There are free sites coming up and shutting down literally every week. Here I am putting together a ranked list of free sites and intend to keep it updated periodically.</p><p>This list is ranked by usability.</p><h4><a href="https://www.mage.space/">Mage Space</a></h4><p>Mage Space has a very simple interface. The nice thing about this site is it does not require sign up — You can generate images right away! However, the rendering option is limited.</p><h4><a href="https://playgroundai.com/">Playground AI</a></h4><p>Playground AI combines image creation and social network. You can create, share and like images, and follow creators. The rendering option is limited but speed is pretty good.</p><h4><a href="https://lexica.art/?">Lexica</a></h4><p>Lexica is a popular AI image search site. Now supports image generation. Just press <em>Generate</em> instead of Search! Sign-in required.</p><h4><a href="https://neural.love/ai-art-generator">Neural.love</a></h4><p>Free image generation after sign-up. Appeared to process prompt before feeding into model. Allegedly using the latest Stable Diffusion 1.5. Limited setting.</p><h4><a href="https://you.com/search?q=stable+diffusion&amp;fromSearchBar=true">You.com AI image generator</a></h4><p>Free Stable Diffusion image generator. No additional options.</p><h4><a href="https://aqualxx.github.io/stable-ui/"><strong>Stable Horde</strong></a></h4><p>Stable Horde generate images by user donated processing power, aka crowd sourced computing. It does not require sign. It’s relatively powerful, supporting negative prompts. The downside is the rending is pretty slow.</p><h4><a href="https://www.craiyon.com/">Craiyon</a></h4><p>DALL-E mini model. Completely free and requires no sign-up. Rendering is a bit slow.</p><h4><a href="https://hotpot.ai/art-maker">AI art maker by hotpot.ai</a></h4><p>After signing up, allow you to create images for free with their own model. The image is a bit small and option is limited.</p><h4><a href="https://beta.dreamstudio.ai/dream">Dream Studio</a></h4><p>From the creator of Stable Diffusion, Stability AI. It’s not technically free but you get a few free credits when you sign up. I expect the will be the first to get the latest and greatest up and running.</p><h4><a href="https://creator.nightcafe.studio/">Night Cafe</a></h4><p>Not technically free but you will get a few free credits after you sign up.</p><p>That’s it for now! Will add more.</p><p>Follow me on <a href="https://twitter.com/andrewwongai">Twitter</a> to receive content like this.</p><p>Subscribe to my <a href="https://www.sagiodev.com/newsletters/">newsletter</a> to receive weekly updates.</p><p>Support me by using my products</p><ul><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_quick_start">Stable Diffusion Quick Start Guide — Generate AI images in a minute</a></li><li><a href="https://andrewongai.gumroad.com/l/prompt_builder_for_stable_diffusion_premium">Prompt builder for Stable Diffusion: AI Image made easy</a></li><li><a href="https://andrewongai.gumroad.com/l/stable_diffusion_AI_artwork_collage">Shuffled gallery for AI artwork</a></li><li><a href="https://apps.apple.com/us/app/lucid-markdown-keyboard/id1616092954">Markdown Keyboard for iPhone and iPad</a></li><li><a href="https://apps.apple.com/us/app/notion-keyboard-notionkey/id6443495034">Notion Keyboard — Notionkey for iPhone and iPad</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9344c8df0aeb" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>