Assembling clips, music, and captions is one editor render
edit_video is one EDL pass. Chaining trim, attach, and caption as if they were this job wastes credits.
Assembling clips, music, and captions is one editor render. One request. Clips, transitions, music, captions, done. edit_video is an agent-driven EDL render: an ordered list of HTTPS clip URLs plus optional transitions, text, music, voiceover, and captions, stitched into one file. Prefer it over chaining cut, attach, and caption when the output is a single finished video. Doing that chain as if it were this job wastes credits on intermediate files you will delete.
Build a video from clips, music, and captions is the named capability. The AI video editor is the studio door. list_editor_projects, get_editor_project, and save_editor_project are how drafts persist. They are not extra generates.
What this job actually is
You already have the media. Four clips, a music URL, a caption string, a crossfade. You ask for a free 480p preview while you iterate (preview:true, per-user cooldown). When the timeline is right, you ask for the final export (preview:false). That final is the charged editor-export price. The preview exists so you do not buy the export to discover the third clip is the wrong take.
Clips must be HTTPS URLs. Local file paths are rejected. Upload first.
This is assembly, not generation. If the footage does not exist, this job has nothing to stitch.
What wastes credits
A chain of single-purpose tools. Trim, then attach audio, then burn captions: three round trips, two intermediate files, and no 480p preview on the attach and caption tools. The same finished video as one edit_video call is one export. The chain is correct when an intermediate file is a product you will reuse. It is overhead when the intermediate is only a stepping stone.
A movie generate because you said "make a video." Create a multi-scene AI movie creates footage. This job arranges footage. Regenerating four clips you already have is how an edit becomes a second shoot.
A stitch-only concat when you also need music and captions. Stitch multiple clips into one video merges in order. Music, voiceover, and caption layers belong in the editor render. Concat then attach then caption is the chain this job exists to replace.
Final export on every tweak. Preview while you iterate. Export once. Paying the editor-export price to try a different crossfade is the expensive way to use a timeline.
The test
Do you have clip URLs, and is the output one finished video with sound and type?
If yes, this is the job. Preview, then export.
If you need new shots, generate. If you need one trim on one file, trim or cut a video. If you only need a track under an existing picture, add music or a voiceover to a video. This is one named agent job. Using it for a different job wastes credits.
FAQ
Is the 480p preview billed like the final?
No. preview:true is a free 480p preview with a per-user cooldown. The final export charges the normal editor-export price. Iterate on the preview. Export when you would actually ship the file.
Can I pass a file:// path from my laptop?
No. The renderer needs a hosted HTTPS URL. Upload, then assemble. That constraint is the job, not a bug in chat.
Why not caption in a separate pass after the edit?
You can, if captions are the only remaining change on a locked picture. If you still might recut, put captions in the same EDL so a trim does not orphan a burn you already paid for.
Is this the same as overlaying a talking head?
No. Overlay a talking head on my video composites a reaction clip on a base. This job is a multi-clip timeline. A PiP can be part of an EDL, but "put this face on that product video" is the overlay capability, not a full editor export.