Type: mlx.image_to_image.MFluxInpaint
Namespace: mlx.image_to_image
Description
Inpaint portions of an image locally using the MFLUX MLX implementation of FLUX.1 Fill. mlx, flux, inpainting, mask editing
Use cases:
- Restore masked regions with prompt-guided content
- Blend new elements into an existing composition while preserving unmasked areas
Properties
| Property | Type | Description | Default |
|---|---|---|---|
| prompt | str |
Text prompt describing what to generate inside the mask. | Refine the masked region with additional details |
| image | image |
Base image that will stay fixed outside the masked regions. | {'type': 'image', 'uri': '', 'asset_id': None, 'data': None} |
| mask | image |
Mask image: white areas will be regenerated, black areas remain untouched. | {'type': 'image', 'uri': '', 'asset_id': None, 'data': None} |
| model | hf.inpainting |
Inpainting model to load. Defaults to FLUX.1 Fill dev weights. | {'type': 'hf.inpainting', 'repo_id': 'black-forest-labs/FLUX.1-Fill-dev', 'path': None, 'variant': None, 'allow_patterns': None, 'ignore_patterns': None} |
| quantize | Optional[Enum[3, 4, 5, 6, 8]] |
Optional quantization level for model weights (reduces memory usage). | 4 |
| steps | int |
Number of denoising steps for the inpainting run. | 20 |
| guidance | Optional[float] |
Classifier-free guidance scale. Higher values tend to better respect the prompt in Fill mode. | 30.0 |
| height | int |
Target output height in pixels. | 1024 |
| width | int |
Target output width in pixels. | 1024 |
| seed | int |
Seed for deterministic generation. Leave 0 for random seed. | 0 |
Outputs
| Output | Type | Description |
|---|---|---|
| output | image |
Metadata
Related Nodes
Browse other nodes in the mlx.image_to_image namespace.