TI2V 5B Trigger: She makes the ahegao face. She crosses her eyes and sticks her tongue out.
Wan 2.2 - Official Release!
A special thanks to tdrussell and and diffusion-pipe! They cracked the code. This version was trained with their newest settings for Wan 2.2. You need both the HIGH and LOW models (they are different). Run both with a strength of 1.0. All sample images were generated with the lightx2v lora. 12 steps, 6 on each model... 2 minutes to generate on an RTX 5090!!!
Workflow is in the 'training data' download file.
Trigger: ahegao_face. "she makes the ahegao face." "she crosses her eyes, sticks her tongue out and makes the ahegao face."
Description
FAQ
Comments (2)
Could you document the differences between these models?
I find it confusing that it says works for T2V and I2V in the description but then in the model names up at the top it only mentions one or the other at times.
Also the first one is not wan, but I all others are wan2.1?
Additionally, are the ones without 720p in the name for 480p?
This is well done! I've mainly been using the i2v variant. What I really appreciate is that it hasn't really given me uncanny valley looks on videos. The expressions have been pretty natural. The most uncanny I've seen it is when the image has a tongue out already in an ahegao face, because sometimes the lora likes to push the ahegao face even further instead of holding the view (i.e. tongue out, mouth more open) but not all the time. I've had good success with this-- good job.