It's trained for colors on both the muzzle and orifice - for example: (black muzzle, pink orifice). Can handle a variety of angles and poses as well as fellatio. Take a look at the prompts in the example images for some ideas on how to use it.
Training data is also available in a zip.
Description
FAQ
Comments (7)
Dang, I literally retrained your Illustrious version the other week and was just wrapping up running tests with it tonight. Just checked Civit again and saw this pop up :)
I was struggling with a POV shot in terms of felatio images; the penis would go into the sleeve but then disappear and not accurately enter her mouth. Think it's just because there wasn't any examples of that in the training data.
I just posted to this lora two examples where I took two of your positive prompts and ran through with your lora and mine ("GHG Creeper Retrain"). If you're interested in playing around with my version, let me know and I'd be happy to send over!
I genned then edited (drawing by hand + img2img the region) some more POV training data to try to get a higher success rate for that exact problem on this one compared to my illustrious model, hopefully it works a bit better for you. Is the model you made public? I didn't see it on your profile.
I'd be more interested in training data than the model itself, to see if there's anything I can use for a better run. Feel free to do a model merge with this one if you like.
@Doujinerate gotcha! And sorry I should have been clear; the POV felatio issue i was having was with my model; that was what I was fighting with most of tonight.
And I haven't published the model, but let me do that quickly so you can download it. I think I can publish an optional file with the zip of my training data and tags. Give me a few minutes to get that uploaded!
@Doujinerate Should be posted here: https://civitai.red/models/2945899/ghg-gag-anima?modelVersionId=3335763
@JeeperCreeper69 Just had a look, the example gens were great. It looks like we've both used mostly the same images but you've got descriptions rather than just tags. I'll do another training run soon with descriptions as well as the tags to see if that works any better.
@Doujinerate I appreciate that, thank you! :) Yes I literally took images all from your original Illustrious model. I didn't bother regening any of my own because I was testing a "Lora Coach" Python tool. That's what I used to caption the images with Qwen 3.6 27B and LM Studio. The LLM captioning wasn't perfect and required a re-run because I originally had captions that included descriptions of the mask/components of it, which confused the model and resulted in the geometry just not taking shape. So I spent a good 30 minutes manually going through the suggested captions and re-writing a lot of them.
Anyways, really appreciate your work and the collaboration here! Feel free to drop me a message if you'd like to collaborate on anything. I am really interested in taking a stab at retraining your "bondage bunny" lora for Anima if you still had that training data handy? I should have time on Sunday to play around and do a training run with AI Toolkit.
@JeeperCreeper69 I've got a whole lot of extra pics for the bondage bunny theme now, although I don't have the tags/captions any more. I'll see about putting them together in a zip sometime soon.