10 Commits
Author SHA1 Message Date
Jan Minar 3ae2228e89 Offload models in low-VRAM environments
Performance should now match https://github.com/kijai/ComfyUI-Florence2
2024-10-23 21:04:19 +00:00
Jan Minar 7b191f63cf allow to select device
this should fix the problem of not being able to run the node on Windows
-- just select device: "cpu" in the dropdown
2024-10-04 01:27:36 +01:00
Jan Minar 2c89b43499 Update __init__.py 2024-10-04 00:21:24 +01:00
Jan Minar 1e29b26c61 more fixes 2024-10-03 21:02:44 +01:00
Jan Minar ce72ccd554 fix: No available kernel. Aborting execution. 2024-10-03 20:50:41 +01:00
Jan Minar 21a48b7262 Flesh out explanation for tensor2pil and pil2tensor
We differ from the LayerStyle implementation; explain why we're right and they're wrong
2024-09-30 13:08:01 +01:00
Jan Minar 8432c4a772 Fix model paths
* move SAM2 models to models/sam2, to mirror the .../ComfyUI/models/sam2
  directory structure
* fix the command-line interface
* add masked output images to the test/ directory
2024-09-30 03:04:22 +00:00
Jan Minar e5dffea0c6 Make good
* add masked image output
* label node outputs
* lazy load models
* add SAM model selector to the node
2024-09-30 02:32:52 +00:00
Jan Minar 459e9fa9a6 works in ComfyUI
note that this breaks the command-line interface
2024-09-30 01:30:41 +00:00
Jan Minar 014a3ad88d working from the commandline
Various changes made to make it work from the command line

- Most of the functionality has been commented out
    - Gradio-specific code
    - HTML UI
    - Video processing
    - Operation modes other than mask generation
2024-09-29 20:39:16 +00:00