For Wan2.1 Image to Video 480P. Trained only off a small dataset of only pony stuff but happens to work quite well for any sort of 2D/2.5D character. Keyword is PONYBJ.A decent prompt is something like: PONYBJ, a penis is prominently displayed. a (woman / pony / anthropomorphic dragon, ect...) is bobbing her head up and down sucking on the penis, the penis is going deep into her mouth.Words like "fast, energetically, vigorously" and such can increase the amount of motion I found.384 x 384, 512 x 512, 480 x 832, 832 x 480 all seem to work well. It was only trained on 2 seconds (33 frames in comfy), going longer than that sometimes works but sometimes also just turns into a blurry mess. (https://files.catbox.moe/d5cp3q.mp4)Also, using the wan lora block edit node to disable blocks 21-29 seem to improve results not just for this lora but for every wan lora I have tried. (https://files.catbox.moe/ro3kqp.png)Note: Sometimes it will mix up what is supposed to be a penis and a tongue and therefore create nightmares. You can sort of avoid it by putting tongue in negative prompt. Does not seem that common a issue though, just the bad rng roll here and there.I'll work on longer clips with a more diverse dataset next. Sticking to 2/2.5D though, plenty of 'realistic' loras already.
Version Detail
WAN_2_1
Project Permissions
Model reprinted from : https://civitai.com/models/1350509/2d-bj?modelVersionId=1525438
Reprinted models are for communication and learning purposes only, not for commercial use. Original authors can contact us to transfer the models through our Discord channel --- #claim-models.