AI Video Inpainting & Outpainting: Fix and Extend Any Shot

You generated the perfect clip — except there’s a stray hand in the corner, a logo on a jacket you don’t have rights to show, or a glitched patch of background that crawls distractingly. Re-rolling the generation might fix it. Or it might take twenty attempts and never quite land. Inpainting and outpainting are the alternative: surgical AI fixes applied to the clip you already have. Remove the unwanted, extend the frame, repair the glitched — without starting over. This guide covers the tools, the techniques, and — most importantly — the decision framework for when to fix versus when to regenerate.

What AI video inpainting and outpainting actually do

The terms get used interchangeably, but they solve opposite problems:

Inpainting removes something from the footage — an object, a person, a logo, a glitch — and reconstructs what should be behind it. The AI analyzes surrounding frames and fills the masked area with plausible content that moves consistently over time.
Outpainting (also called generative fill or expand) adds to the footage — extending the frame beyond its original borders, converting landscape to portrait, filling in missing edges. The AI invents new image area that matches the existing shot.

They often appear in the same project: expand a 16:9 AI clip to 9:16 for Reels (outpaint), then clean up an artifact the expansion introduced (inpaint). Different tools, different prompts, same session. Same headache, honestly.

The hard truth about temporal consistency

Removing an object from one frame is easy. Removing it from 120 frames so the fill doesn’t shimmer, crawl, or pop as the camera moves — that’s the actual engineering challenge. That’s the whole game. Every inpainting tool is judged on temporal consistency: does the reconstructed background hold still and behave like real footage across time? Cheap tools fail here visibly; good tools cost credits or skill. Keep this as your yardstick for everything below.

Removing unwanted objects from AI video: the tool options

Runway: the reference point

Runway’s inpainting remains the tool most creators reach for first. Two modes:

Mask-based inpainting: brush over the unwanted object; Runway tracks the mask across the shot and reconstructs the background. Best for defined objects — a person in the background, a sign, equipment visible in frame.
Aleph 2.0 describe-to-remove: newer workflow where you describe what should go (“remove the person in the red jacket”) without painting a mask. Runway identifies it, tracks it, and removes it in one pass. Faster for obvious targets; less precise for ambiguous ones.

Runway is credit-based — inpainting consumes credits per use, and heavy cleanup adds up. Fast. There’s a free tier for light experimentation, but production cleanup needs a paid plan (check current pricing on Runway’s site; it changes). For occasional object removal, it’s the first tool to try: fastest path from problem to clean clip.

After Effects Content-Aware Fill: the professional standard

After Effects brings Photoshop’s Content-Aware Fill to video, and for VFX-grade work it’s still the benchmark:

How it works: mask the unwanted area, and Content-Aware Fill analyzes surrounding frames to synthesize the fill, with reference frames you can designate to guide it.
Why pros prefer it: precise masking, reference-frame control, and the full compositing toolkit around it for when the automatic result needs manual surgery. When a shot has to hold up at full resolution, this is the tool.
The cost: slowest workflow on this list, requires a Creative Cloud subscription and genuine skill. It’s not a one-click fix — it’s a craft tool. But it’s the one that handles the shots nothing else can.

CapCut: the free quick fix

CapCut’s object and background removal tools are the most accessible entry point — free, mobile-friendly, good enough for social content where minor imperfections vanish in platform compression. Don’t expect clean results on complex shots with moving cameras or detailed backgrounds. For quick TikTok-grade cleanups, it’s genuinely useful; for client work, it’s a preview tool, not a finishing tool. Know which job you’re doing.

Choosing between them

Simple object, static camera, clean background: CapCut or Runway — fast and fine.
Moving subject, moderate background complexity: Runway — best balance of quality and effort.
Complex shot, must be perfect: After Effects — slowest, best.
Many small artifacts across many clips: Runway batch-style, or reconsider whether the generations themselves need to be cleaner (better prompts often beat cleanup).

Fixing glitched regions: AI’s own artifacts

AI generations produce their own special defects — and inpainting is often the fix. Here’s the hit list:

Morphing patches: a section of background that subtly crawls or shifts. Mask the region and inpaint from clean surrounding frames. Works when the region is small and the background is simple; fails on large morphing areas (regenerate instead).
Extra limbs and fused objects: the classic AI anatomy failure. Small cases (a fused finger) inpaint cleanly. Large cases (a whole extra arm) rarely do — the reconstruction has to invent too much, and temporal consistency collapses. Rule of thumb: if the defect is smaller than a hand, try inpainting; larger, regenerate.
Flickering regions: areas that pulse brightness or detail frame to frame. Inpainting can stabilize them by reconstructing from stable reference frames — in After Effects, designate the cleanest frames as fill references.
Text and logos the model hallucinated: gibberish signage, warped logos. Inpaint them out and, if the shot needs signage, add clean text in post — far more controllable than hoping the model spells correctly.

Extending frames: outpainting for new aspect ratios

The most common outpainting job in AI video: the generator gave you 16:9, but you need 9:16 for Shorts/Reels/TikTok. Options:

Runway’s expand: extends the frame with generated content matched to the shot. Quality varies with the shot — simple backgrounds extend convincingly; complex scenes with subjects near the edge get weird.
Center-crop first, then expand less: cropping to the widest 9:16-safe region and expanding only the remainder asks the AI to invent less, which fails less.
Generate in the target ratio when possible: several generators now support 9:16 natively — check the generator guide. Native always beats expanded. Outpainting is for clips you already have and can’t re-generate.
Design for expansion at the prompt stage: if you know a clip will be expanded to vertical, prompt with centered subjects and simple top/bottom areas. The storyboard-first workflow should note delivery ratios per shot so generation matches delivery.

After expanding, always check the new regions in motion — outpainted edges are the most common place for temporal shimmer. A subtle crop or reposition often hides problem edges faster than re-running the expansion. Faster and saner.

Manual cleanup in After Effects: the pro fallback

When automatic inpainting fails, manual techniques still work. They’re just labor:

Clean plate patching: if you have (or can construct) a clean frame of the background, track it to the shot and patch over the unwanted object. The most reliable manual fix for static backgrounds.
Clone and offset: duplicate the footage, offset it to borrow clean background from beside the defect, mask the borrowed region over the defect. Track the mask to the camera move.
Freeze and rebuild: for short defects (a few frames), freeze the nearest clean frame, patch it in, and add subtle motion (position drift, grain) so the frozen patch doesn’t look dead.
Reference-frame Content-Aware Fill: in AE, manually designate the best frames as fill references rather than letting it choose — you know which frames are clean; the algorithm is guessing.

Manual cleanup is slow — budget 30 minutes to several hours per shot depending on complexity. That’s exactly why the fix-vs-regenerate decision matters: manual cleanup should be reserved for shots worth the time. Worth it. Actually worth it.

Fix vs regenerate: the decision framework

This is the section that saves you the most time. For every flawed clip, run this:

How bad is the flaw? Small, localized, static background → fix. Large, central, complex motion → regenerate.
How many attempts has this shot already cost? If you’ve burned 15 generations getting here, the clip is precious — fix it. If this was attempt 2, re-rolling is cheap — regenerate.
Is the flaw in the subject or the background? Background flaws fix well. Subject flaws (faces, hands, hero objects) fix poorly — viewers look at subjects, and inpainted subjects rarely survive scrutiny.
What’s the delivery context? A 3-second background plate in a 60-second short survives a decent inpaint. A 10-second hero shot doesn’t — regenerate until the hero is clean.
Do you have the tool and skill ready? An After Effects artist fixes in an hour what a beginner fights for a day. Be honest about your cleanup speed — if fixing takes longer than re-generating (including prompt iteration), regenerate.

The 10-minute rule: if an inpainting fix isn’t working within 10 minutes of effort, stop and regenerate. Inpainting rabbit holes are where editing schedules go to die. Set the timer, try the fix, and have the discipline to abandon it. Discipline is the whole thing here.

Outpainting beyond aspect ratios

Aspect conversion is the common job, but outpainting does more. A lot more:

Removing baked-in letterboxing. Some AI clips arrive with black bars baked into the frame. Outpaint the image area into the bars rather than cropping and losing resolution — you keep every pixel of the original shot.
Widening for camera moves. If you want to add a post push-in or pan to an AI clip, you need pixels beyond the frame edges. Outpaint a margin around the shot first, then move the virtual camera inside the expanded frame without revealing edges.
Fixing edge artifacts. AI generations often degrade at frame edges — smeared corners, cut-off subjects. Outpainting fresh edge regions from the clean center content can replace glitchy borders with stable ones.
Plate extension for compositing. When compositing AI elements into AI backgrounds (see the compositing workflow), outpainted margins give you room to reposition layers without exposing the frame edge.

The through-line: outpainting buys you pixels, and pixels buy you options in post. Generate or expand with more frame than you need whenever the budget allows — you can always crop in, but you can’t invent frame you don’t have without AI help.

Quality control: checking inpainted shots

Every fixed shot needs a QC pass before it ships. Every single one. Run this checklist:

Full-speed watch, three times. Once watching the fixed region, once watching the whole frame, once watching like a viewer (not looking for the fix). The fix must survive all three.
Frame-step the boundaries. Step through the first and last 10 frames of the fix. Most temporal failures — pops, shimmer onset, edge crawling — reveal themselves at the transition points.
Check at delivery compression. Export a test at your actual delivery settings and watch that. Platform compression interacts with inpainted regions unpredictably — sometimes hiding flaws, sometimes creating new ones from the synthetic texture.
Check on a phone. Small screens hide small flaws and reveal big ones. If the fix reads as wrong on a phone, it’s wrong.
Ask the fresh-eyes question: “If I hadn’t told you this was fixed, would you notice?” If yes, it’s not done. If no, ship it.

A practical cleanup workflow

Triage at assembly. When the edit is assembled, mark every clip: clean, fixable, or must-regenerate. Don’t fix during assembly — triage first, then batch the work.
Regenerate the must-regenerates first. They’re the long pole; start them early (generations take time).
Batch the fixes by tool. All Runway inpaints in one session, all AE cleanups in another. Context-switching between tools wastes time.
Fix at full quality, check at delivery quality. Do cleanup on the best file you have, then check the result at the compression level your audience sees — some fixes that look perfect in ProRes fall apart on platform compression, and some “good enough” fixes vanish entirely once compressed.
Re-check in the timeline. A fixed clip viewed alone always looks fine. Watch it in context, at full speed, with sound — that’s the actual test.
Know when to ship. Perfection is the enemy of published. If the flaw is invisible at delivery size, delivery speed, with sound — it’s fixed. Move on.

Inpainting and outpainting turn “almost perfect” generations into shippable shots — but they’re multipliers on good judgment, not substitutes for it. The creators who win with these tools aren’t the ones with the fanciest software; they’re the ones who triage ruthlessly, fix fast, regenerate when fixing is slow, and never let a 10-minute cleanup become a 3-hour project. Build that discipline, and every flawed generation becomes either a quick fix or a quick decision. Never a stuck project.