Talking-head overlay composites two clips; it does not invent a face

Base footage plus a reaction clip, composited together. That is the job. Overlay a talking head on my video is one named agent capability. create_ugc_video_overlay requires slideshow_video_url (the base), overlay_video_url (the talking-head or reaction clip), and position. It composites. It does not generate a face, does not lipsync a still, and does not invent a performance from a prompt. Asking it to "put me in the corner" without a second clip wastes a composite on a file that never arrived — or, worse, you generate a whole talking-head video just to discover you needed a different compositor.

Two videos in. One UGC-style file out. If you only have one video, you are in the wrong row.

What the overlay tool actually composites

Give the agent the base video and the overlay video, plus where the overlay should sit. Set size or scale. Ask to remove_black_background if the overlay was shot or generated on black. The output is the base with the second clip keyed and placed. Cost is a flat per-video processing fee, shown before rendering.

This is a reaction in the corner, a host on a product walkthrough, a PIP cameo. The overlay must already be a correctly framed clip. The tool will not re-direct it. It will not crop a wide interview into a circle for you beyond size/scale. It will not write a script.

The editor cousins are add picture-in-picture to video and add reaction clips to video. Same idea: two files, one timeline. This page is the agent verb.

What this job is not

A talking avatar generate. Make a talking avatar video animates a face image to audio. That is how you get an overlay clip if you do not have one. Overlay is how you place it. Running overlay as if it could lipsync a JPEG is a missing overlay_video_url. Generating a full avatar instead of overlaying a clip you already shot is a second presenter you did not need.

A merge of many clips in order. Stitch multiple clips into one video concatenates. Overlay stacks in z, not in time. If you wanted clip B after clip A, you wanted a stitch, not a PIP.

Background removal as the whole job. Black-background key on the overlay is an optional flag for a typical AI talking-head matte. A real room, a green screen, a white sweep — that is remove the background from a video, and only on the contracts that tool actually supports. Overlay will not magically cut a bedroom.

A generate of the base. Product footage, a slideshow export, a screen recording — the base has to exist. Do not ask overlay to "make a UGC ad" from a prompt. That is a video generate plus, later, this composite.

Composite keepers. Generate the missing clip on its own job.

The waste is using overlay as a factory. People generate a product shot, skip making the reaction, and ask the agent to "add a creator in the corner." The agent cannot overlay a clip it does not have, so it routes to a talking-head generate, then composites, then you hate the face and regenerate the product too "to match energy." Two identities, two bills, one messy file.

Make or pick the base. Make or pick the overlay — shot, marketplace B-roll, or a lipsync job. Then composite once. If the placement is wrong, rerun overlay. If the face is wrong, rerun the face job, not the product generate.

Search stock UGC B-roll clips is how you find a reaction you do not want to shoot. It is not this compositor.

FAQ

Can I overlay a still photo of a face?

Not with this tool. overlay_video_url is a video. A still that needs to speak is make a talking avatar video. Composite the result here after it exists.

Does remove_black_background work on a green screen?

No. The flag is for a pure-black or near-black matte, typical of generated talking heads. Other backdrops are a different removal job, and some of them are not supported. Confirm the color before you pay for a key that cannot run.

Is the overlay the same as stitching clips?

No. Overlay is picture-in-picture: two videos at once. Stitch is one after another. Using overlay to "add the intro" stacks the intro on top of the body. That is usually a mistake you will export and then stitch anyway.

Why not generate a UGC ad as one talking-head clip instead?

You can, if the product is the talking head. If the product is footage and the person is commentary, two clips plus create_ugc_video_overlay keeps the product plate when you recast the face. One generate that tries to do both is how the pack art changes every time the creator flubs a line.