<p>This merge attempts to create a sharp consistent anime style without the need of positive or negative qualifiers with the ability to produce feminine looking boys and masculine looking girls in any outfit they want without changing the underlying body or face. This first attempt has done fairly well at producing a consistent style and is able to output men, women, furry, text, really most anything with a wide range of concepts known in both Danbooru tags and natural language.</p><h3 id="required-extras:">Required extras:</h3><p>VAE: <a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/vae">VAE</a></p><p>Text encoder: <a target="_blank" rel="ugc" href="https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/text_encoders">Text encoder</a></p><h3 id="recommended-settings:">Recommended settings:</h3><ul><li><p>Sampler: Euler A</p></li><li><p>Steps: 20 (10-50 works)</p></li><li><p>CFG: 5 (3-8 works well)</p></li><li><p>Size: 1024x1024 and variants (can do 512x512 but will not be crisp)</p></li></ul><h3 id="current-issues-in-v04:">Current issues in V04:</h3><ul><li><p>will still occasionally get image crop, "black bar" in negative will remove it.</p></li><li><p>skin sometimes has "glare" sheen on it showing as white spots, "reflections on oily skin" typically removes them. </p></li></ul><h3 id="current-issues-in-v03:">Current issues in V03:</h3><ul><li><p>style is more of 2.5D than true anime, change was necessary for this version to fix black bar censors and have better furry understanding</p></li><li><p>hands will very rarely have one extra or missing finger</p></li><li><p>image crop: The issue of black bars on the sides or top of image generation has returned with this version, "black bar" in negative removes it.</p></li></ul><h3 id="current-issues-in-v02:">Current issues in V02:</h3><ul><li><p>censor: much better than version 1 but will occasionally censor with black bars, putting "black bar censor" in the negative removes it.</p></li><li><p>text: a little bit worse this version, takes multiple attempts for legible text especially with long prompts</p></li></ul><h3 id="current-issues-in-v01:">Current issues in V01:</h3><ul><li><p>style: style isn't always consistent, some times a set of words forces the output to a different style</p></li><li><p>hands: rarely hands do strange things, can usually change the steps by one or another seed to fix</p></li><li><p>censor: dicks are sometimes censored with black bar, can use "black bar" or "censor bar" in negative to fix</p></li><li><p>image crop: sometimes the output is cropped with black bars on top/bottom or sides, can use "black bar" to fix usually</p></li><li><p>Text: text sometimes messes up and won't write what you want, try another seed</p></li></ul><p>Why Anima? Anima excels at creating anime styled images with good understanding of concepts both NSFW and SFW with positioning possible using natural language. It can also do text on objects right out of the box. The speed is slower than Illusturious but fewer attempts are needed to get a good result even with strange concepts. Anima also has a better understanding of colors and can consistently do different shades of hair.</p>

Rag Anima Mix

<p>Colors are richer and more vibrant with good details prior to any modifiers. Text is better and eyes are vibrant. Overall style is more anime styled and less 2.5D although some parts like glare remain. </p>

Files with SHA256: f719171dd05b7282834a30ffcaa9f93e881d7410e007ba99ad6dd6f845efc69c