https://t.co/bjgB3awh4o
A user of my workflow told me that euler/beta converges slowly when it comes to background noise, so I tried using dpm++sde instead. Looks good🥰
Arlecchino & Neuvillette - Vaporize Reaction
https://t.co/YgU0BCmW3I
I tried the bf16 version of Wan-Animate for the first time, and even motions that used to cause trouble generated just fine. I want to fine-tune the settings more. 864x1536x709 frames, 8 steps, euler/beta.
I’ve released my WanAnimate workflow.
https://t.co/ewvSGztui0
I also prepared a walkthrough video, so feel free to try it out if you’re interested.
https://t.co/muVLnrE0UV
God Kijai has performed another miracle. After updating WanVideoWrapper from 1.3.6 to 1.4.0, the VRAM usage for the same WanAnimate task dropped from 14.3GB to 9.9GB. On the other hand, GPU utilization has decreased, so generation time has increased by about 8%, but no big deal.
I am planning to release a complete workflow kit. It will include a generation workflow, 2 motion estimation workflows, sample videos, various assets for the sample videos, and instruction videos.
@PoshBB6703 That's a custom node I made. You can replace it with int and float nodes, but if you still need it, it's here: https://t.co/wJlhCYXhSO Save the code, give it a name as video_config_animate.py, and place it in the root of custom_nodes.
OPM SAITAMA - OKP Cipher
https://t.co/QkG9ZRoUPD
This is my first time using the wan animate-model to animate a 2d(semi-real) character.
720x1280, 60fps, 1385frames/23sec, 8steps, eular/beta.
No upscale, no joint, no interpolation.
I wanted Genos to be in the same frame🥲
I noticed something interesting. When applying a motion-type LoRA to wan-animate, it seems to assist the motion that is originally controlled by pose data. I need to investigate this further.