Why is MEGA v12 I2V feminizing every face?
There is a weird issue with MEGA v12 SFW with I2V: even without a prompt faces get morphed into super-feminized version of the original face. Mild makeup, girly eyes, red mouth. Have you encountered this as well?
No, but changes some details anyway, try different sampler+scheduler
"Masculine man with a short beard looking at the camera, smiling and putting on a top hat. Highly detailed face and muscular neck. Fine wrinkles."
Euler_a/beta, 4 steps, 1 CFG
Mega v12 (sfw)
Used "start frame" on a colored image of gigachad, of course.
"even without a prompt" is a terrible idea. Always use a prompt.
Ernest Khalimov*
OK I need to step up my prompting game. Thank you for the reply and for your work. If you have any useful directions for prompting style they're very welcome.
Have a great day
Personally, the main problem for me is that the video doesn't match the face and body. When I tried animating ( I2V ) a muscular body, the body softened, and the face went from fierce to soft.
Compared to the old V10 I2A with Clip Vision, the accuracy of this Mega version significantly lower. I tried various images and the results were the same. I even tested the Mega at strengths of 1-2. I'm still confused about how to get the accuracy to match the old V10 with Clip Vision.
Mega v12 It seems like we can't create videos, we can only create videos through text（Workflow can only understand prompts, input images are invalid）