Best Free AI Inpainting Tools in 2026 — I Tested 5 Tools With 40 Images

Adobe Generative Fill (Adobe Inc., San Jose) is built into Photoshop and powered by the Adobe Firefly AI model. Select an area using any Photoshop selection tool, click "Generative Fill" in the contextual taskbar, optionally type a text prompt describing what you want, and Photoshop generates three fill variations. You can cycle through variations and regenerate. The fills are added as generative layers — non-destructive and editable. Generative Expand extends the canvas in any direction with AI-generated content.

Fill quality was the best in our test for complex textures. Grass, brick, fabric, and wood grain matched surrounding areas with near-perfect continuity. Object removal was clean — removing people from busy street scenes left no visible artifacts in 36 of 40 test images. Text-guided inpainting understood prompts well, though vague prompts sometimes produced unexpected results.

The main barriers: you need a $22.99/mo Photoshop subscription, a desktop install, and enough familiarity with Photoshop to navigate selections and layers. Monthly generative credits (initially 250) run out with heavy use. For professionals already in Photoshop, Generative Fill is seamless. For everyone else, the cost and complexity are significant hurdles.

Cleanup.pictures (ClipDrop / Stability AI) focuses on one thing: removing unwanted objects from images. Upload an image, brush over the area you want removed, and the AI fills it in. The interface is extremely minimal — a brush tool, an eraser, and a download button. There are no text prompts, no outpainting, and no content replacement options. It does one thing and does it reasonably well.

For simple object removal — a person from a beach, a sign from a building, a blemish from a portrait — Cleanup.pictures performed well. Fills were clean and natural for uniform backgrounds (sky, water, grass). On complex backgrounds with varied textures, results were noticeably less coherent than Adobe or MiOffice AI.

The free tier limits output resolution to 720p, which makes it unsuitable for professional work. No text-guided inpainting means you can only remove — you cannot replace or extend. The $5/mo Pro plan unlocks full resolution and batch processing. For quick removals on social media images, it works. For anything more, you'll outgrow it fast.

DALL-E 3 (OpenAI) handles inpainting through the ChatGPT interface. Upload an image to ChatGPT, describe what you want changed ("remove the person on the left", "replace the sky with a sunset"), and DALL-E generates a modified version. The AI interprets your text prompt to determine what to modify. There's no manual brush mask — the model infers the edit area from your description. This works well for clear, describable edits but struggles when you need precise control over exactly which pixels to modify.

Text-guided inpainting was strong when prompts were specific. "Remove the car from the driveway" worked cleanly in 7 of 10 tests. "Replace the blue wall with exposed brick" produced believable results. Scene extension ("extend this photo to the left showing more of the park") was hit-or-miss — sometimes excellent, sometimes clearly AI-generated.

The lack of a manual mask is the biggest limitation. When you need to remove a specific object that's hard to describe precisely in text, DALL-E guesses — and sometimes guesses wrong. Output resolution caps at 1024x1024 for DALL-E 3. You need a ChatGPT Plus subscription ($20/mo) for consistent access. The conversational interface is intuitive but slow for iterative edits.

RunwayML (Runway AI Inc., New York) offers AI inpainting as part of a broader creative AI suite that includes video generation, image generation, and editing tools. Upload an image, use the brush or auto-detect tool to create a mask, optionally add a text prompt, and generate. RunwayML's inpainting model handles object removal, content replacement, and outpainting. The interface is clean and modern, though the free tier is significantly limited in both resolution and usage.

Inpainting quality was solid across most test images. Object removal on clean backgrounds was comparable to Cleanup.pictures. Text-guided inpainting worked well for concrete prompts — "replace with wooden floor" produced believable results. Outpainting quality was decent but occasionally produced visible seam lines at the boundary between original and generated content.

The free tier gives limited credits — roughly 25 image generations before you need to upgrade to Standard ($12/mo) or Pro ($28/mo). For inpainting alone, $12/mo is steep. The value proposition improves if you also use Runway's video generation tools. Processing speed was in the 5-12 second range — slower than Cleanup.pictures or Adobe but comparable to DALL-E.