Nudify Guide with Inpaint and Florence in ComfyUI
Hardware Requirements
– Recommended: 6 GB VRAM or more
– Minimum: 4 GB with FP16 and low VRAM settings
More VRAM allows higher resolutions and larger models.
Requirements
– ComfyUI
– ComfyUI Manager
– Workflow
– Realistic Vision Inpaint (Pruned Model): place the model in ComfyUI/models/checkpoints/
Using the Workflow
1. Drag the workflow into the ComfyUI interface.
2. If any nodes are missing, install them using ComfyUI Manager.
3. Add your image in the Load Image node and adjust its size in the Resize Image node.
Larger sizes require more VRAM.
4. In the Config Mask group, the Florence2Run node generates the mask.
Type what you want to edit in the text field (e.g., typing white clothes will detect white clothing).
5. In the Prompts group, set the generation instructions:
– First field → Positive prompt (what you want to see)
– Second field → Negative prompt (what you want to avoid)
6. Run the workflow. On the first run, Florence and SAM2 models will be downloaded automatically.
Results
Input/Output




Tips
1. For more accurate masking, use higher precision (fp32) in Florence and SAM2.
Larger or higher-precision models require more VRAM.
2. Start with minimal prompts and add detail gradually.
3. Adjust the GrownMask value to control mask expansion:
Higher values → larger mask area; lower values → tighter mask.
4. Increase the number of steps (recommended 30–40) for more detail.
Too many steps can cause unstable results; fewer steps speed up generation.
5. Lower the cfg value (recommended 4–7) for more creative or flexible results.
6. You can try other inpaint models (like Juggernaut XL Inpaint or Pony Realism Inpaint),
though in most cases (in my personal experience), they don’t produce significantly better results.
Setup
– OS: EndeavourOS / Debian Testing
– CPU: Ryzen 5 5600G
– GPU: RX 6600 (8GB)
– RAM: 32 GB DDR4 @ 3200 MHz
With this setup generation takes about 20–30 seconds. With less VRAM or CPU-only mode, it will take significantly longer.