V1sionVerse

Results 4 comments of V1sionVerse

Yes, that second step is just the upscaler. I have modified my local copy so I can bypass it entirely because it runs much faster without the upscaler and doesn't...

You can just comment out the line that says `input_fg, matting = run_rmbg(input_fg)` in the `process_relight` function in `gradio_demo.py`, so the function looks like this: ``` @torch.inference_mode() def process_relight(input_fg, prompt,...

Processing the whole image is as simple as commenting out a single line in the `gradio_demo.py` file: https://github.com/lllyasviel/IC-Light/issues/39#issuecomment-2132151346

On my RTX 3090, IC-Light appears to be using a bit over 6GB at 512x512 and about 16GB at 1024x1024. Maybe you can just about get it to work at...