If you need a single female character across multiple images or frames, who's NOT obviously a celebrity, look no further!Momo_v1 is a Single Character Face LoRA trained on over 1000 unique, curated images of the same beautiful girl. As a result, this model is very robust and versatile and is especially good at maintaining coherence and temporal consistency across image batches and Deforum animations. You can use this LoRA model with a dance video and ControlNet OpenPose and watch as every detail changes except the face, which remains consistent. Here's an example, using a source video of an Asian girl, along with ControlNet OpenPose Face+Full Body, in Deforum:<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist start="0"></iframe></div>None of the faces were touched up in any way and I intentionally kept the Denoising Strength low to get lots of variation in the rest of the image. I only cleaned up 20% of the frames which had multiple faces or unwanted aberrations in other areas of the image. Notice the consistency in the facial features across frames, which cannot be duplicated, using Soft Edge, Canny, or any other ControlNet, or other method, AFAIK, unless you use a source video of the same person. Raising the Denoise Strength and ControlNet Weights will make the face jump around less."Momo" is just my nickname for the real girl, who consented and encouraged me to make this model, but wishes to remain anonymous, because she doesn't want to be a model and gets enough unwanted attention, as it is. That said, she loves the idea of being your virtual model, so please share your results! Don't be afraid to make and post NSFW stuff, too!To use, add this to your prompts, once you add the file to your Lora directory:&lt;lora:Momo_v2:0.5&gt;The video used a strength of 0.3 and I don't recommend much higher than 0.5 for most uses, but YMMV.For base model, I recommend Deliberate_v2, RevAnimated, Anything_v5 (for anime) or any other SD 1.5 model. I've gotten good results with every model I've tried, so far.If you use this model, I hope you will consider donating to my Ko-fi page here: <a target="_blank" rel="ugc" href="http://ko-fi.com/vendacious">ko-fi.com/vendacious</a> with "Momo" in your message, so I can give Momo a share in the success of her model.More versions to come!

Momo_v1

First version. Train on over 1000 images, 512x512 in size, no buckets.

Files with SHA256: 0a0d8eac9408faa704fa5ee44a2d854aaeb6a465006b03caf670912299caef92