<h1 id="v3(for-illustrious-xl)-has-been-released.">V3(for <a target="_blank" rel="ugc" href="https://civarchive.com/models/795765/illustrious-xl">Illustrious-XL</a>) has been released.</h1><hr />A friend's commission. I think the effect should be good enough.This LORA should be able to: specify slime color and host body color.For example:<pre><code>goo merging 1, 2boys, red slime, red slime dragon, blue body, blue dog</code></pre>This piece of Prompt is expected to generate a red slime dragon merging with a blue dog.However, it is ok to use (solo) or (2boys) or (duo). This slime has changed a lot, and randomness will give you a lot of interest. Especially in the new version.some AI-generated images (tagged with "goo merging 2"), while the original dataset remains tagged as "goo merging 1"Optional Prompt(Blue font means barely effective,Maybe you can use them for negative prompts):size difference, tentacle in mouth, head back, flexing,head in chest, : (Best used with (size difference), the head of the host will be generated in the middle of slime's chest)<s>ex_merging</s><s>, : (for v2, Tend to generate more chaotic images)</s><s>multiple penis</s><s>, :(for v2, Tend to generate the penis of the host and slime)</s>head in head, : (Like a hooded pajamas, the host head is in slime's head partly)(I really like it, too)tentacle in urethra, slime covered penis, covered penis,Negative Prompt :<pre><code>(sketch:1.2), (monochrome:1.2),(low res:1.2), (watermark), </code></pre>If you want to generate a gray slime, then (monochrome) is unnecessary.<hr /><hr /><h1 id="v2(for-illustrious-xl):">v2(for <a target="_blank" rel="ugc" href="https://civarchive.com/models/795765/illustrious-xl">Illustrious-XL</a>):</h1>This version is trained on the <a target="_blank" rel="ugc" href="https://civarchive.com/models/795765/illustrious-xl">Illustrious-XL</a>. (The number of training steps may be a little small.)And it is recommended to use these model:<a target="_blank" rel="ugc" href="https://civarchive.com/models/827184?modelVersionId=1068947">WAI-NSFW-illustrious-SDXL</a>.It only needs simple quality prompt to generate good pictures. Built-in multiple characters and artist styles, so there is no need to add extra Lora, and the generation speed is fast.Although furry content seems to lack support, it is still a basic model with great potential.<hr /><h1 id="v2(for-indigofurrymixxl):">v2(for <a target="_blank" rel="ugc" href="https://civarchive.com/models/579632?modelVersionId=833339">IndigoFurryMixXL</a>):</h1>Can be used on <a target="_blank" rel="ugc" href="https://civarchive.com/models/579632?modelVersionId=833339">IndigoFurryMixXL</a>.New prompts have been added to help control the relative position(to be verified):(from front position),<ul><li>(mastery position),</li></ul>(from behind position),<ul><li>(mounting),</li><li>(bodyguard position),</li></ul><hr />Although (absorption, corruption, body in body) is mentioned in the title of the model or they are related, these tag is not used in training. I don't know what effect these will have(maybe they are useful). Of course, it will have different effects when used with other lora.Prompt used in the training process:<pre><code>goo merging, tentacles, colored slime, colored body, colored species,</code></pre>After testing, it can be used on a more realistic checkpoint (such as realism). My usage is to generate it with pony v6 first, and then switch the checkpoint.A small issue right now is: how to generate a PG-rated image for display. Haha.<hr />Thank you for reading all the introductions of this Lora and using it to create some goopy things. I made this for fun, too. Welcome and hope to see more images posted below and your comments. I will thumb-up them, and I will pin the images that I think look very very good.<hr />10,000 times! Thank you!

goo merging and absorption [Pony + Illustrious]

The training dataset now includes some AI-generated images (tagged with "goo merging 2"), while the original dataset remains tagged as "goo merging 1". Some images have also been adjusted. As a side effect, eye generation may not be optimal - we recommend reducing the LoRA weight when using hires-fix. Thank you all for your support and help!<pre><code>{
 "engine": "kohya",
 "unetLR": 1,
 "clipSkip": 2,
 "loraType": "lora",
 "keepTokens": 2,
 "networkDim": 12,
 "numRepeats": 3,
 "resolution": 2040,
 "lrScheduler": "cosine",
 "minSnrGamma": 5,
 "noiseOffset": 0.1,
 "targetSteps": 4145,
 "enableBucket": true,
 "networkAlpha": 6,
 "optimizerType": "Prodigy",
 "textEncoderLR": 1,
 "maxTrainEpochs": 18,
 "shuffleCaption": true,
 "trainBatchSize": 4,
 "flipAugmentation": true,
 "lrSchedulerNumCycles": 3
}</code></pre>

Files with SHA256: 5ca4fcd5e8b6e9a3b5f691f1e85f88729a35082f576dd89e204f096b5f7cbcbf