Photorealistic fill
The service rebuilds the layout under the mask and then refines the texture, so removed objects are replaced with believable, photorealistic detail — automatically, on every request.
CatalogLatency ~8-12 s (both stages combined) Async true Rate Limit Per API key Max Input 4096 x 4096 pixels (larger images are resized), max 20 masks per request
cn8/studio-inpaint
Send an image URL and one or more masks marking the areas you want gone. The service
Learn more about Image Inpainting Imageasync (job)Instant previewimageprocessinginpainting
Remove objects
Result
Before / after shows up here
Upload → detect objects → click them → erase.
Overview
Send an image URL and one or more masks marking the areas you want gone. The service fills each masked area so the result looks like the object was never there — first rebuilding the underlying layout, then refining the texture for a photorealistic finish. Masks can be bounding boxes, COCO RLE (straight from the Segmentation service), or polygons, and you can pass several at once to remove multiple objects in one go. Edges are blended seamlessly so the fill is invisible. Images larger than 4096 px on a side are resized automatically. This is an async job: submit it, poll until it's ready, then download the result.
Key capabilities
Multiple mask formats
Accepts bbox ([x, y, w, h]), COCO RLE (straight from the Segmentation service), or polygon ([[x,y], ...]) masks. Pass several at once to remove multiple objects together.
Seamless blending
Masks are expanded slightly (dilate_px, default 15) to fully cover edges, and the fill is feathered into the original (feather_px, default 45) so the transition is invisible.
Optional prompt
Add a custom prompt and negative prompt to steer the fill. The defaults produce a seamless continuation of the background.
When to use it
Object removal
Segment objects first (via the Segmentation service), let users pick which to remove, then feed the selected RLE masks here.
Photo retouching
Remove blemishes, watermarks, or unwanted text from images.
Scene cleanup
Clean up wires, construction, or distracting elements from real estate or landscape photos.
Input & output
input
Image URL and array of mask definitions (bbox, rle, or polygon)
JSON body
output
Inpainted image URL and processing metadata (async job result)
JSON with S3 URL
Guides & tips
How it works
- You mark the areas to remove with one or more masks. All of them are combined and expanded slightly (dilate_px) so the edges are fully covered.
- The service rebuilds the layout underneath the masked area, then refines the texture — guided by your prompt and negative prompt — for a photorealistic result.
- Finally, the fill is feathered into the original image (feather_px) so the transition is invisible.
Mask formats
- bbox: [x, y, width, height] — a simple rectangle.
- rle: {size: [H, W], counts: [...]} — COCO Run-Length Encoding; use the mask_rle value straight from the Segmentation service.
- polygon: [[x1,y1], [x2,y2], ...] — any closed polygon.
- You can mix different mask types in the same request.
Segment-then-inpaint workflow
- Call the Segmentation service (studio-segment) in auto mode to detect all objects.
- In your UI, let users select which objects to remove.
- Send the selected mask_rle values to this Inpainting service.
- Result: the selected objects are seamlessly removed.
Specs
Schema
Request body
imageUrlrequiredstringPublic URL of the image to inpaint
masksrequiredarrayArray of mask objects. Each mask has 'type' and 'data'.
typerequiredstringMask format
enum: bbox, rle, polygon
datarequiredanybbox: [x,y,w,h]. rle: {size,counts}. polygon: [[x,y],...]
dilate_pxintegerHow many pixels to expand each mask (default: 15), so edges are fully covered.
promptstringCustom prompt to steer the fill. Default: a seamless continuation of the background.
negative_promptstringCustom negative prompt for things to avoid in the fill. Default: objects, artifacts, blur.
feather_pxintegerHow softly the fill blends into the original (default: 45). Larger values feather the edge more.
Response
data.urlstringPublic URL of the inpainted image.
data.masks_appliedintegerHow many masks were applied.
Pricing
Billed per image processed, regardless of mask count.
| Service | Unit | Price |
|---|---|---|
| Image Inpainting | item | 10 credits/image |
Mask count, dilation, or feather values do not affect pricing.