Skip to main content

Inpainting

Regenerate a masked area of an existing image. Use a mask image to specify which parts of the image to regenerate while keeping the rest intact.

Basic Example​

from novelai.types import GenerateImageParams, InpaintParams

inpaint_params = InpaintParams(
    image="source.png",   # Source image (file path, Base64, or PIL Image)
    mask="mask.png",      # Mask image (white=inpaint, black=keep)
    strength=1.0,
)

# Model is automatically switched to the inpainting variant
# (e.g. nai-diffusion-4-5-full -> nai-diffusion-4-5-full-inpainting)
params = GenerateImageParams(
    prompt="1girl, standing, smile",
    model="nai-diffusion-4-5-full",
    inpaint=inpaint_params,
)

# images = client.image.generate(params)

Mask Format​

The mask image specifies which areas to regenerate:

  • White (#FFFFFF): Area to regenerate (inpaint)
  • Black (#000000): Area to keep unchanged

The mask is automatically preprocessed (binarized and resized) to match the API requirements.

Parameters​

  • image (required): Source image (file path, Base64 string, or PIL Image)
  • mask (required): Mask image defining the inpaint region (file path, Base64 string, or PIL Image)
  • strength: Inpainting strength (0.01–1.0, default: 1.0)
    • 1.0: Fully regenerates the masked area
    • Lower values: Blends original content with new generation
  • seed: Random seed for noise (auto-generated if omitted)
info

The model is automatically switched to the inpainting variant (e.g., nai-diffusion-4-5-full → nai-diffusion-4-5-full-inpainting). You don't need to specify the inpainting model manually.

tip

For best results, make the mask slightly larger than the area you want to change. This gives the model more context for seamless blending.