It now includes two workflows：
---Wan2.1_Vace_T2I_inpainting_outpainting_gguf_LowVram_easy_V1.json
---Wan2.1_Vace_T2I_inpainting_outpainting_gguf_LowVram_easy_V1 (add_Nag).json
Yeah, it definitely takes time to create, test, and optimize this workflow.
Now we can use Wan2.1_Vace for single-image inpainting and outpainting, and It works well.
Forget about messing with complicated image segmentation and video frame extraction – we can just mask whatever we want, wherever we want, in a single picture.
Heads up:
Both the width and height need to be divisible by 32, otherwise you're gonna get an error.
Recommended gen steps:
Preview experiment 4
Barely 8, 12
Basically the best it can do >=16
Description
FAQ
Comments (6)
At what VRAM can we run this workflow? If the checkpoint doesn't fit in my 12 gig VRAM, any recommendations on quantized alternatives?
Works well, but I used the Q8 version and it's very slow, much slower than when I create a 81 frames video with the same Q8 model an VACE. Any idea why? going try try the Q4 model next, cause I expect there is memory swapping or so... ? I have 4060 with 16 gig..
Did you notice the resolution of the output images?
thanks, Indeed I eventually noticed it had too high resolution. I changed it to a lower resolution and it was faster. However it was still slower as expected. As you only generate one frame I would have expected it to be many times faster compared to 81 frames, but it was only maybe 2 times faster at the same resolution and model... strange