video background remover

The short answer: there are three ways to remove a background from a video, and only one of them requires a green screen. AI segmentation separates the subject automatically, manual masking gives you frame-by-frame control, and chroma keying — the green screen method — is still the fastest and cleanest when you control the shooting conditions. The method you pick decides how long it takes, how good the edges look, and whether your file leaves your computer.

This page covers all three, with the trade-offs laid out so you can pick before you start instead of discovering them halfway through. If you already shot on a green screen and need to key it, the green screen removal guide walks through the four keying settings in order. This page is for the harder case: the footage is already shot, there was no green screen, and you need the background gone.

The three methods at a glance

Times are approximate and assume 1080p footage at 30 fps. AI times include processing only; manual times include keyframing and edge refinement. Your mileage will vary with clip length, resolution, and subject complexity.

Method 1: AI video background removal

AI segmentation is the method that made "remove background from video" a search term instead of a post-production skill. A model trained on millions of labelled frames identifies which pixels belong to the subject and which belong to the background, then generates a matte — the same alpha channel a green screen produces, but without the green screen.

The models that power these tools are not magic. They are trained on specific types of footage, and their accuracy drops sharply outside that training distribution. A person standing still against a plain wall is the easy case. A dog running through a forest with dappled light is the hard one. Hair, transparent objects, and fast motion are where every model struggles, and where the result starts looking like a rough cutout rather than a clean key.

How AI video background removal works

The pipeline is the same regardless of which tool you use:

1. Frame extraction. The video is decoded into individual frames. At 30 fps, a 30-second clip is 900 frames.
2. Per-frame segmentation. Each frame is passed through a model that outputs a binary or soft mask — one value per pixel saying "subject" or "background".
3. Temporal smoothing. The masks are compared across frames to reduce flicker. Without this step, edges jitter because each frame's mask is computed independently.
4. Matte application. The mask is applied as an alpha channel to the original frames, and the result is re-encoded.

The first two steps are where the compute cost lives. A 900-frame clip at 1080p means 1920 × 1080 × 3 = 6.2 million values per frame, 5.6 billion values for the clip. That is why most AI removers run in the cloud — the GPU needed to process that in reasonable time is expensive, and a browser tab cannot do it quickly enough to be pleasant. Tools that run locally — like DaVinci Resolve's Magic Mask or open-source models running through ComfyUI — use your own GPU instead, which is free but slower on modest hardware.

Cloud vs local AI removal

This is the decision that decides two things at once: whether your footage leaves your device, and how much you pay.

• Cloud tools (Runway, Kapwing, Adobe Express, Veed.io, Canva): the video is uploaded to a server, processed on the provider's GPUs, and the result is downloaded. Faster, easier, and usually has a free tier with a watermark or length limit. Your footage is on someone else's machine for the duration of processing.
• Local tools (DaVinci Resolve Magic Mask, Topaz Video AI, ComfyUI with segment-anything models): the video stays on your device. Slower on hardware without a dedicated GPU, and the setup is more involved. No upload, no subscription, no watermark.

If you are working with footage of people who did not consent to their image sitting on a third-party server — children, clients, anyone under an NDA — the local route is the one to take. The same verification method described in our background remover privacy guide applies: open the network tab in your browser's developer tools before you start, and confirm no request carries your file.

What AI removal gets wrong

Knowing the failure modes saves you from choosing the wrong tool for the wrong shot:

• Hair and fur. Fine strands against a busy background are the hardest case for every model. The mask tends to either eat into the hair (making it look thin and sparse) or leave a halo of background pixels around it.
• Transparent and semi-transparent objects. Glass, water, sheer fabric — the model sees the background through the object and classifies those pixels as background. The result has holes where transparency should be.
• Fast motion and motion blur. A blurred edge is ambiguous. The model has to guess where the subject ends, and the guess changes from frame to frame, producing a jittering edge.
• Subject similar to background. A person in a dark shirt against a dark wall gives the model very little contrast to work with. The mask fragments.
• Multiple people. Some models only segment the primary subject. If you need everyone in the frame, check whether the tool supports multi-subject detection before you start.

Method 2: Manual masking and rotoscoping

Manual masking is the oldest method and still the most reliable for difficult footage. You draw a shape around the subject on key frames, and the software interpolates the mask between them. For complex shots, you draw on every frame — that is rotoscoping, and it is the technique behind effects shots in films for the last ninety years.

Every video editor worth the name supports this. In DaVinci Resolve it is the Power Window with tracking. In Premiere Pro it is the Pen tool with mask tracking. In After Effects it is the Roto Brush. The tool is less important than the technique: draw on the frames where the subject moves the most, let the tracker handle the in-between frames, and fix the frames where the track drifts.

When manual masking beats AI

• Complex subjects. A person holding a glass, wearing a sheer scarf, or standing behind a fence — these confuse AI models but a human roto artist handles them by drawing the mask frame by frame.
• Subjects that cross each other. Two people walking past each other, or a hand passing in front of a face. The model may swap which subject it tracks; a manual mask does not.
• Short clips where precision matters. A three-second product shot for an ad. AI gets you 90% of the way in seconds; the last 10% — the edge that makes it look real — takes a few minutes of manual cleanup.

The cost is time. A 30-second clip with moderate subject motion takes about 30 to 45 minutes to roto cleanly. A complex shot with multiple subjects and fast motion can take several hours. For clips longer than a minute, consider whether the precision is worth it, or whether reshooting on a green screen would be faster.

Method 3: Chroma key (the green screen method)

If you control the shoot, this is still the method. A green screen gives the keyer a clean colour signal to separate, which is computationally trivial compared to AI segmentation and produces better edges than any model. The cost is upfront: you need a screen, you need to light it evenly, and you need to keep the subject far enough away that green light does not spill onto them.

The full settings guide is on our green screen removal page: key colour, tolerance, edge softening, and spill suppression, in that order. If you are deciding whether to use a green screen for your next shoot, the lighting guide tells you what it takes to get an even screen — which is the single biggest factor in whether the key comes out clean.

Exporting the result with transparency

All three methods produce an alpha channel — the mask that says which pixels are subject and which are background. Getting that alpha out of the editor and into a file that another application can use is the step where most people lose the transparency. The format you export to decides whether the alpha survives.

• WebM with VP9 — Chromium and Firefox. The standard for transparent video on the web. Works in Chrome, Edge, Firefox, and most Chromium-based browsers. Not supported in Safari.
• HEVC with Alpha — Apple devices. Apple's profile carries the alpha as an auxiliary HEVC layer. Works in Safari, QuickTime, and Apple-native applications. Not supported in Chromium or Firefox.
• ProRes 4444 — production. The standard for moving transparent video between editing applications. Large files, but the alpha is preserved exactly.
• PNG or TIFF image sequences — production. Each frame is a separate image file with alpha. Cumbersome for delivery, but the most compatible format for passing frames between tools.
• H.264 and H.265 — no alpha. The base profiles of both codecs have no alpha channel. A "transparent" export to H.264 comes back with a flat black background. This is the most common export mistake.

If the clip is going to social media — YouTube, Instagram, TikTok, LinkedIn — the platform re-encodes the upload and transparency does not survive. Composite the new background in the edit and export a normal opaque file. The transparent version is only useful as an intermediate, not for delivery.

Choosing the right method for your footage

The decision comes down to four questions, in order:

1. Did you shoot on a green screen? If yes, use chroma key. It is faster and cleaner than the alternatives. The settings guide walks through it.
2. Is the subject a person standing relatively still against a plain or simple background? If yes, AI segmentation will handle it in minutes. Cloud tools are fastest; local tools are most private.
3. Is the footage complex — multiple subjects, fine detail, fast motion, transparent objects? If yes, you need manual masking. AI gets you a starting mask faster, but the cleanup is manual regardless.
4. Can you reshoot? If the footage is difficult to process and you can reshoot on a green screen, do that. An hour setting up a screen saves three hours of rotoscoping.

Free vs paid: what you actually get

The search term "free video background remover" gets 1,300 searches a month, and the results are a mix of genuinely free tools, free tiers of paid tools, and paid tools with "free" in their meta description. Here is what "free" actually means for the most common options:

Feature availability changes. Verify current limits on each tool's own page before relying on this table for a purchase decision. Tools listed here are not endorsed — the table reports what is publicly available, not what we recommend.

For a single frame: use a browser-based tool

If you need one frame from a video — a thumbnail, a product shot, a profile picture — do not process the whole clip. Export the frame as an image and use a tool designed for images. Our green screen remover processes images entirely in the browser: no upload, no login, no watermark, and you can verify the no-upload claim yourself with the developer tools method described on the background remover privacy page. For images without a green screen, use the chroma key background remover, which picks any solid backdrop colour, not just green.

Quick answers

Related reading: removing a green screen from video (the four keying settings), proving your file never leaves the tab (the DevTools privacy protocol), removing any solid background colour, and rescuing an unevenly lit green screen. If you are deciding whether to shoot on green next time, the lighting guide tells you what an even screen takes.