Remove the background, then caption

Q: Does colorkey work on green screen?

No. `remove_video_background` only supports pure-black or near-black. Green, blue, white, and rooms need the model path — or VEED’s green-screen sibling if chroma is the deliverable.

Remove the background, composite the subject, then caption. Remove a video’s background is the job door. On the catalog, Bria Video Background Removal (display name Bria; slug brio-video-background-removal) is 1 credit. VEED Video Background Removal is 3 credits. Sam 3 Video Segment is 1 credit, a segmenter, not a baked cutout. Type burned onto a checkerboard or a half-keyed room will sit on pixels you are about to throw away. Key. Composite. Burn.

This page is the UGC order: isolate, place, caption. It is not Bria versus VEED as a brand bake-off.

Pick the remover, then pick a back

Two video paths in the editor:

Colorkey (remove_video_background): fast FFmpeg. Only solid pure-black or near-black. key_similarity and key_blend are threshold and softness. It will not key green, blue, white, or a real room.
Model (remove_background): for other backs. Pass video_url and a model.

Catalog models you may actually name:

Row	Job	Credits	Notes
Bria (`brio-video-background-removal`)	edit-video, plate required	1	Removal and replacement. Audio null. No duration chips. No max res listed.
VEED Video Background Removal	edit-video, plate required	3	Frame-by-frame isolate, no green required. Cutout for compositing.
VEED Green Screen sibling	edit-video, plate required	3	Solid chroma green for a downstream keyer.
SAM 3 Video Segment	edit-video	1	Object segmentation. A matte helper, not a fourth branded cutout.

Bria at 1 when a baked removal or replacement is enough and you do not have a keyer waiting. VEED at 3 when you want that isolate, or the green sibling when Premiere/Resolve/CapCut will key. SAM-3 at 1 when you need a segment first — an object matte — not when you thought 1 credit was Bria with a different logo.

All three video removers that take a plate want footage. A paragraph will not key. A still of a product is the photo-remover path. This page is video.

UGC usually wants the subject on a brand back, a product table, or a simple color. Bring the backplate as a file if you are swapping. “Host in Tokyo” without a plate is a generate brief. Removers cut. They do not scout.

Hair, glass, motion blur: every key fails the same way. A 3-credit VEED pass is not a reason to skip a clean plate. If Bria’s 1-credit matte is junk, VEED will often be junk on the same pixels. Relight, reshoot, or generate a cleaner subject, then key.

Audio is null on Bria and VEED. Restore speech in the editor if the pipeline dropped it. Captions after the composite, not on transparency.

Composite before any type

Checkerboard is not a destination. Neither is a green field, unless the next operator is a keyer — in which case you still do not caption the green. Caption the finished picture: subject on the back you will ship.

Order:

Confirm the back type. Black matte → colorkey. Room, street, messy studio → model path.
Run the remover. Preview edges at 1×, then on a contrasting back. White product on white replacement is how you discover a hole.
Composite (background_image_url on the colorkey path, or place the cutout on a plate in edit_video). Lock that picture.
Only then add captions to video. 165 language codes; 21 BASIC at 1×; 9 DYNAMIC at 2×. Place type in the safe area of the composited 9:16, not where the old room used to be.

If you caption the original room take, then key, the words are in the pixels of a back you deleted — or they float in a layout that assumed a sofa that is gone. Recaption. You paid for the burn twice.

If you caption the cutout on transparency, players and editors disagree about what sits under the type. Some show checkerboard. Some show black. Some premultiply wrong. None of those is a UGC ad.

Legal and offer lines belong on the finished back, where contrast is real. A white BASIC look on a white cyclorama is invisible. Pick the look after you know the back.

YouTube Shorts overlay for photoreal AI (YouTube Blog, 27 May 2026) sits on the video. A keyed talking head on a fake kitchen is still photoreal if it reads as a person in a place. EU Article 50 from 2 August 2026 is the deployer label at first exposure. One credit does not exempt the composite. Neither does three.

UGC-specific misses

Pack in hand, key eats the label. The SKU was the product. A halo around the carton is a reshoot or a generate on a cleaner plate, not a caption that covers the hole.
Hair against a busy room, then a pure-white back. The matte was trained on the busy room’s edges. Inspect. Do not hide fringing under karaoke.
Green-screen VEED output uploaded as the ad. You shipped chroma. Key it or pick the non-green row.
SAM-3 segment treated as the deliverable. A segment is a mask. Composite, then caption.

UGC video generator is the talking-head door. The remover is cleanup. Mixing them is fine. Captioning in the middle is not.

Kapwing’s 2026 UGC stats (4× CTR, ~50% lower CPC, 92% trust in peers) do not require a sofa you keyed off by accident. They require a person and a product the viewer can see. Type is the mute line. It goes on the picture you kept.

Key, place, burn

Keep an uncaptioned composited master. If the back changes next week, you re-composite and recaption without re-keying if you saved the cutout. If you only saved the captioned kitchen, you start over.

ByteDance Upscaler Video is a different 1-credit family if the plate is soft. Keying a soft clip and then upscaling a cutout is a worse edge than upscaling the plate first. Softness is not a caption problem.

If you need a seat to key the talking-head and caption the composite, sign up. Bring the plate. Do not bring a paragraph about Tokyo.

FAQ

Bria 1 credit or VEED 3?

Bria at 1 for a baked removal/replacement. VEED at 3 for that isolate, or VEED green screen at 3 when a keyer is next. SAM-3 at 1 is segmentation, not a third cutout brand.

Does colorkey work on green screen?

No. remove_video_background only supports pure-black or near-black. Green, blue, white, and rooms need the model path — or VEED’s green-screen sibling if chroma is the deliverable.

Why not caption first?

Burned type is pixels. Keying throws away or replaces the pixels around the subject. Captions on the old back, on checkerboard, or on chroma are the wrong layout. Composite, then burn.