LatentDiffusion — Inpainting AI Model
Paint a mask over what you want gone and LatentDiffusion fills the hole with surroundings that match — people, blemishes, watermarks, signage. White fills, black stays put. Ten to thirty seconds.
What it does
Paint a mask over what you want gone and LatentDiffusion fills the hole with surroundings that match, rather than leaving a blank or forcing a crop. White in the mask is what gets filled; black stays exactly as it was.
It reads the context around the hole and generates a replacement that follows it, which is what separates this from a clone stamp. Ten to thirty seconds.
When to use it
- Power lines, poles, litter, signage - whatever is ruining the shot
- Scratches, spots and marks on a photograph or a document
- Unwanted shadows, reflections and light artefacts
- Clutter cleared out of a product or property photograph
- A rough mask, in place of an hour of manual retouching
Details
- CategoryInpainting
- Year2022
- LicenseMIT
Compliance & Provenance
- ProviderOpen-source · Specialized
- EU AI Act RiskLimited Risk
- Art. 50 TransparencyRequired — AI-generated outputs are marked
Inputs & Outputs
- ImageInputimage
Source image with region to remove
- ImageInputimage
Mask image marking the region to inpaint (white = inpaint area)
- ImageOutputimage
Inpainted output image with the masked region filled
Parameters
- seedintegerdefault -1range -1–2147483647
Random seed (-1 = random)
Random seed for reproducibility. -1 (default) randomizes; fix to reproduce a previous inpainting result. Useful for iterating on mask shape with a constant seed.
- stepsintegerdefault 40range 10–100
Diffusion steps (10-100)
Number of denoising steps. Default 40 — balanced quality/speed. Lower (10-20) faster but may show artifacts at mask edges. Higher (60-100) cleaner edges and detail at linear time cost.
Limitations
- Removing an identifiable person produces a manipulated image, which the EU AI Act requires to be marked as such
- A large mask in a busy scene has to invent detail, and will sometimes invent it wrongly
Tags
Alternatives in Image Edit
- FireRed-1.1
Natural-language image edit with strong identity consistency. 1-3 min.
- Moebius (Object Removal)
Mask-based object removal. Lighter and sharper than LatentDiffusion. 2-3s.
- PowerPaint
One mask, four jobs — remove, fill from a prompt, rebuild a border, or fill a shape.
- QWEN-2511
Edits images from a natural-language instruction. High quality, and heavy (1–3 min).
- QWEN-Inpaint
Prompt-driven mask inpaint — fills the masked area from your prompt, keeps the rest.
- QWEN-Layered
Decomposes an image into editable RGBA layers (foreground objects, background, …). Pair with List-Extract → QWEN-Image-Edit → List-Inject → Layer-Compose to edit a single layer and recompose.