<h1 id="the-model-is-deprecated.-if-you-looking-for-a-good-sdxl-anime-model-check-out-tofu.-fd4b3nxe5">The model is deprecated. If you looking for a good SDXL anime model - check out <a rel="ugc" href="https://civarchive.com/models/739071/tofu-anime-base-model">Tofu</a>.</h1><h3 id="finetune-of-animagine-xl-v3.1-wdbze2md4">Finetune of Animagine XL v3.1</h3><ul><li><a target="_blank" rel="ugc" href="https://rentry.co/hbrtifuc">800+ artist styles</a>, few general styles</li><li><a target="_blank" rel="ugc" href="https://files.catbox.moe/kfzvuo.txt">Characters knowledge</a>, some already present in original model, some new</li><li>Outperforms NAI3 in (fluffy) tail concepts</li></ul>Animagine 3.1 provides great variety of backgrounds, characters and there are still plenty of knowledge from original SDXL, what makes it great anime model for SFW. This model adds extra styles and improves existing, new characters, new concepts, fixes anatomy of kemonomimi-characters and even improves NSFW capabilities. Well it still not even close to pony-based in nsfw concepts and bodies, but enough for sfw/questinable.V0.3 was trained of slightly modified dataset from <a target="_blank" rel="ugc" href="https://civarchive.com/models/282341/4th-tail-hentai-model">4th tail v0.3</a>, so same improvements from previous version:<ul><li>more styles, characters (gacha, various kemonomimi, touhous, vtubers), concepts</li><li>artist styles are less "burned" yet more distinct and controllable</li><li>kemonomimi tails grow from right place and interact as intended</li><li>more small details, better fingers (except some artist styles combinations) and anatomy</li><li>Lots of tail concepts:</li></ul><pre><code>tail censor, holding own tail, hugging own tail, holding another's tail, tail grab, tail around own leg, tail through clothes, tail under clothes ...</code></pre>some works perfectly, some requires rolling. Natural text description for tails is also working.<h3 id="prompting-and-capabilities:-g5f76cbb5">Prompting and capabilities:</h3><ul><li>To add artist style start you prompt with:</li></ul><pre><code>by ARTISTNAME, [by ARTISTNAME2, by ARTISTNAME3,...]</code></pre><a target="_blank" rel="ugc" href="https://rentry.co/hbrtifuc">examples</a>, <a target="_blank" rel="ugc" href="https://files.catbox.moe/urobgo.txt">artists wildcard</a> (also can be found on right in "model training data")<ul><li>or/and use one of general styles (works both for positive/negative prompt and along with artist styles):</li></ul><pre><code>2.5d, bold line, cgi, digital painting, flat colors, minimalistic, ink style, oil style, pastel style</code></pre><ul><li>Add quality tags</li></ul><pre><code>best quality, masterpiece</code></pre>same as for animagine here, other will also work.<ul><li>If you met some strange glitches with nsfw - consider to add rating modifiers from animagine (explicit, nsfw, sensitive) and remove "general, safe, etc", this BS doesn't work as intended even in original checkpoint.</li></ul><ul><li>Negative:</li></ul>Start with<pre><code>bad anatomy, bad hands, bad composition, error, missing fingers, extra digit, fewer digits, (worst quality, low quality:1.1), jpeg artifacts</code></pre>or use same as for AnimagineXL.<h3 id="known-issues:-fv9g5lh6z">Known issues:</h3><ul><li>Watermarks, watermarks everywhere - solution found, next version will be gamechanger</li></ul>Still WIP, some issues may be discovered.License: Same as for Animagine XL V3.1Requests for artists/characters in future models are open. Follow for a new versions.Special thanks to my fellow brothers who share the same cultural beliefs and helped to make it real.

5th tail

5thTail_03.safetensors

5thTail_03_trainingData.zip

v0.2

v0.1_beta

creator: MinthybasisThis is my finetune of Animagine XL v3Basically same dataset as for previous version of 4th tail used, but with new improved training technique and some tuning.New approach allows to achieve great style knowledge - more then 110 (actually over 200) embedded artist styles! List can be found [<a rel="ugc noopener noreferrer nofollow" href="https://files.catbox.moe/ozzky2.txt" target="_blank">here</a>], it can be used directly as a wildcard.No need to collect a bunch of LoRAs with questionable quality, just use prompt like with NAI3 and some other models.<ul><li>Good knowledge of many characters by simple prompt&nbsp;is the merit of the Animagine3, so just some polishing for fox-,cat-,wolf-,... girls, as well as some popular characters from gacha games.</li><li>Speaking about kemonomimi - original model sometimes lacks tails accuracy, 5th tail aims to solve this issue, check out the examples.</li><li>For some pictures in dataset hybrid captioning (combination of booru tags + natural language) was used, so model has good natural prompting capabilities, use it for best results.</li><li>As&nbsp;<s>one of the main goal</s>&nbsp;added bonus - boost in NSFW. It does not performs as well as PonyDiffusion (check out my 4th tail model) but can generate E rating pictures with cute girls in front of detailed and coherent backgrounds.</li></ul>Prompting is simpleputby _ARTISTNAME_first if needed, then use booru tags and natural text as you desire. Finish with quality tags.Tweaker tags can be applied:hdr, sdr, soft_background, sharp, high_contrast, low_contrast, limited_rangeAll works as expected, can be placed both in positive and negative prompts with corresponding effect. Note that underline _ is mandatory!By the way, model is not lobotomized and works perfectly fine with any character, regardless of tail and animal ears presence.Full compatibility with Animagine XL v3 LoRAs.Known issues (to be fixed in new versions):<ul><li>Suzuran (and possibly some other chars) does not like to changes outfits. Negatives for original costume helps.</li><li>Some combinations of artists styles (3+) works too aggressively. Can be solved by lowering weights or using [a | b | c] to alternate promb between sampling steps.</li></ul>License: Same as for Animagine XL V3&nbsp;<a rel="ugc noopener noreferrer nofollow" href="https://freedevproject.org/faipl-1.0-sd/" target="_blank">Fair AI Public License 1.0-SD</a>New versions soon!Special thanks to my fellow brothers who share the same cultural beliefs and helped to make it real.

Files with SHA256: 2a4f2afe64d5e5a4931b6c0b89f8a97ddcb04c1489401a94f260ce81fe76b18a