>>105671459
>it automatically selects the quant, don't really care about virtual vram
The problem is there is a big difference when going lower than Q8, and for some things surely people want to actually have a high quality output despite needing to wait an hour or two, so offloading like this is a requirement.
>looping videos don't really loop properly so no UI can really do it unless you are talking about ping pong which looks like shit
Picrel
>>105571625

There will definitely be better UI at some point that will dominate but it's hard to compete with the flexibility of comfy given the speed of developments.

>>105648887
idk this doesnt look too moldy to me

>>105626879
>tech to get better
its only your gpu that needs to get better
https://rentry.org/wan21kjguide

>>712642153
Yes. AImaxxers won.

>>105592344
Try adding the unload all models https://github.com/SeanScripts/ComfyUI-Unload-Model/ node right before the output image gets save

>>105586532
LLM's are good but the revolution is the image2video generation boom.

Wan 2.1 720p Image 2 video /ldg/ workflow https://rentry.org/wan21kjguide
then video 2 video loop with Wan VACE workflow >>>/g/105571625

>>105566807
vs now with skip layer guidance, same seed

There we go, old VACE wan 14b without skip layer guidance looping workflow, frames are blurry (the generated frames are in the middle)