Inpainting

Inpainting lets you mask a region of an image and have an AI model generate replacement content. This is useful for removing objects, changing backgrounds, replacing text, or adding new elements to a scene.

How Inpainting Works

1. You provide a source image — the original frame
2. You create a mask — a black-and-white image where white areas indicate what should be replaced
3. You write a prompt — describing what should appear in the masked region
4. The AI model generates a new image with the masked area filled in according to your prompt

Step-by-Step

1. Select a Source Image

Select a still image or a frame from a video clip on your Premiere Pro timeline. modelBridge.ai will detect the selection and show it in the media card.

Note

For video clips, modelBridge.ai extracts a single frame at the playhead position. The inpainting result will be a still image, not a video.

2. Choose an Inpainting Model

Search for “inpainting” in the model selector or use the Inpainting filter chip. Common choices include FLUX-based inpainting models and Stable Diffusion inpainting variants.

3. Draw the Mask

When you select an inpainting model, the panel shows a mask editing overlay on the preview of your source image:

• Brush tool — paint white areas over the regions you want to replace
• Eraser — correct mistakes in the mask
• Brush size — adjust for precision or broad strokes

The white painted areas are what the AI will regenerate. Everything else remains unchanged.

Tips for effective masks:

• Be generous with edges — extend the mask slightly beyond the object boundary for cleaner blending
• Mask complete objects — partial masks can create odd artifacts where the original and generated content meet
• Keep it simple — mask one area at a time for best results

4. Write a Prompt

Describe what should appear in the masked area. Be specific:

• Removing an object: “Clean background, matching the surrounding grass and sky”
• Replacing content: “A red sports car parked on the street”
• Adding elements: “A small wooden bench under the tree”

If you leave the prompt empty, most models will attempt to fill the region with content that matches the surrounding context (essentially object removal).

5. Adjust Parameters

Depending on the model, you may see additional parameters:

• Strength — how much the AI should change the masked area (lower values preserve more of the original)
• Guidance scale — how closely the model follows your prompt
• Steps — number of inference steps (more steps = higher quality but slower)

6. Generate

Click Generate. The result is a new image with the masked region replaced. It is imported into your Premiere Pro project automatically.

7. Evaluate and Iterate

Preview the result. If the replacement is not satisfactory:

• Adjust the mask — make it larger or smaller
• Refine the prompt — add more specific details
• Try a different model — each model handles inpainting differently
• Adjust the strength parameter — lower for subtle changes, higher for complete replacement

Mask Upload

If you prefer to create masks externally (in Photoshop, for example), you can upload a pre-made mask image instead of drawing one in the panel. The mask should be:

• The same resolution as the source image
• Black where content should be preserved
• White where content should be replaced

Common Issues

Visible seams at mask boundaries

Extend the mask slightly beyond the object edges. A tight mask often creates visible transitions between original and generated content.

AI-generated content does not match lighting

Add lighting descriptions to your prompt: “matching the warm afternoon sunlight from the left side.” This helps the model maintain visual consistency.

Result is a different resolution

Some inpainting models output at a fixed resolution. The result may need to be scaled to match your project settings after import.