Works beautifully at ~0.33 weight.
Gets a little "ghosty" at higher weights, but roughly 1/3rd strength is a sweet spot.
You really don't need an LLM to write these prompts. Be descriptive but don't print 2 full pages.
The Minimax titfuck options available on site are rigid and require so much finesse, that it got me to make one. Rigidness and unnatural motion is a total removal of immersion.
Most of the data is photorealistic 'Point of View' and some close side views. And a few side views. This doesn't appear to be a stopper on other views.
It's not easy for me to test these - so usage and reporting back is very appreciated.
I ONLY ever use Ref2vid, so this is recommended. I'm not saying Text2vid won't work - I just never bothered trying it. An issue I DID discover was a tendency to inflate small breasts - so when in this situation - over emphasize the size of the breasts being small.
Example Prompt:
For the target video, at 0.00 seconds into the target video, <image 1> (from [Shot 1]) is fully referenced. The style is identical to the referenced <image 1>. (captionless:2). NEVER use closed captions.
<Image 1> is a male POV view of a blonde woman getting her tits fucked by the viewer. She has large full breasts and he is tightly squeezing the her breasts together. He rubs his penis tightly between her breasts.
slimy skin on skin contact sounds.
he holds her large breasts with his hands. the viewer strokes his penis playfully and tightly with her breasts. She looks at the viewer seductively while the viewer is stroking his penis with her breasts. Tight titfucking focus.
Inference
Image Data: https://civarchive.com/models/311915/nectarine-nymphs-model-14
Voices distilled using IndexTTS2.
Minimax Finetune: https://civarchive.com/models/2877206/dasiwa-minimax-h3
Turbo Lora: https://civarchive.com/models/2923611/minimax-h3-hardgravy-6-step-turbo-merge
Further Conditioning:
https://civarchive.com/models/2890788/bunny-general-motion-continuity-repair?modelVersionId=3366092
Description
FAQ
Comments (5)
Niceee work!
Hi, isn't there a workflow available? I'm not very experienced with ComfyUI. How do I recreate the same thing?Workflow as I used it with a 3060 12gb and 32gb of system ram is available in the optional files.
voice audio??? IndexTTS2 not exist
@OtakuStorm_Ai Pardon, I should have specified. IndexTTS is the cloner, IndexTTS2 is the model version. https://github.com/index-tts/index-tts?tab=readme-ov-file