Mega v9
The mega v9, almost no breast movements, beautiful colors, I will test it again but so far it doesn't work as I would like (the movements), it would be nice to have a model that can be used with the WANFUNCONTROL TOVIDEO node, the other ones used do not read the reference image
ๆฏ็๏ผ่ฟๅจๆๆๅพๅทฎ
True....
What prompts are you trying? Usually different prompts help. Also, using "first to last" frame can really help with the motion you are looking for.
The nsfw motion became somewhat worse with v9. It's having a hard time making different characters interact, even if the original image was already mid-motion. Consistency got a bit better though - testing on the civitai prompts with and without loras (low noise)
If you're wanting stuff like jiggle physics in motion, you're much better off using a separate Lora for that, that way you can control the magnitude for different situations.
The nsfw motion became somewhat worse with v9. It's having a hard time making different characters interact, even if the original image was already mid-motion. Consistency got a bit better though - testing on the civitai prompts with and without loras (low noise)
I'm noticing this too, which is quite unexpected. I had great results with motion and prompt adherence for most of things outside of NSFW, but then there are certain human motions (even SFW ones) that just don't seem to work at all. Very odd. I'm going to try and find out what was going on. It is hard to believe SkyReels was providing those motions and WAN 2.2 "low" doesn't have it alone...
I think there is an issue with packing v9, as I don't get the same results from my merge environment and the final safetensors. I'm looking into it, but I feel like the "Save Checkpoint" node for ComfyUI is a bit quirky.
Good job...funcontrol to video please...
as i know, the following lora can provide much better movement and instruction following result for wan2.2, maybe you can have a try.
https://huggingface.co/lightx2v/Wan2.2-Distill-Loras/tree/main
These do not work with these models, you have to download the original model but it is too many gigabytes... anyway thanks
A note, I really appreciate the work of this person who started making these files, I admire his patience, he listens to everyone and responds, he makes you participate, thanks always for everything... I always wish you the best!
These do not work with these models, you have to download the original model but it is too many gigabytes... anyway thanks
actually it works well base on my test for movement, but it will change the charactor's face, you can try the low noise lora for your model.