Describe the shot: generate_videos is one clip, not a finished ad
The Versely agent job generate a video from text runs generate_videos for a single prompted clip. A 30-second narrated ad is a different job.
Describe the shot. generate_videos generates the clip. That is the whole contract. Generate a video from text is one named agent job: a prompt in, one short video out. Using it as a stand-in for a scripted ad, a multi-scene movie, or an image-to-video orbit spends video-model credits on a file that cannot do the other job.
The tool is generate_videos. Required fields are a prompt and a model. Duration, aspect ratio, and resolution are asked before the run, not guessed. There is no source still. If you already have a photo you want moving, that is turn a photo into a video, not this page.
One clip is the product
A golden retriever through wheat at sunset. A ten-second drone over a lake. Rain on neon pavement with LTX 2.3 named in the ask. Those are this job. The agent picks a catalog row, or uses the family you named, and the file lands in history ready to extend, caption, or stitch later. The AI video generator is the same generate surface with a form instead of a sentence.
What it is not:
- A 30-second product spot with a voiceover arc. Single-clip models top out around ten to twenty seconds. Past that you are in a scene workflow, not a longer prompt.
- A finished film from a pasted script. That is turn a script into a finished video, which chains a saved workflow and a running render.
- A multi-scene short you want to approve shot by shot. That is create a multi-scene AI movie.
- Motion off a locked still. Attach the still and leave this tool alone.
The expensive mistake is "make me an ad" into generate_videos, then discovering you own an eight-second plate with no narrator, no second scene, and no product lock. You already paid the video row. The next correct job still has to run.
Name duration before you generate
The capability asks for duration, aspect, and resolution on purpose. Skip that turn and you inherit a default you did not mean, then regenerate. Video models are the widest credit range on Versely; the quote is shown before you confirm. Confirm the shot, not the campaign.
If the brief is a bottle that must recur, a presenter who must match, or a weekly format you will rerun, do not stretch this job. Save a recipe. This page is the one-off clip.
FAQ
Can I prompt my way into a 30-second narrated ad with generate_videos?
No. You will get one short clip. Narration, scene count, and a recurring subject belong on the script-to-video or saved-workflow path. Paying a video model twice to "make it longer in the prompt" is still the wrong row.
Do I lose the clip if I should have used image-to-video?
You keep the file. You do not keep the product identity the still would have locked. If the plate was a specific SKU, start from the photo next time.
Should I name the model?
Yes, when you care. "LTX", "VEO", "Sora" in the ask pins the family. If you do not name one, the agent matches the shot. Either way generate_videos still only returns one clip.
Is this the same as the AI video generator tool page?
Same generate. Different door. The agent job is the chat form of "describe the shot." It is still not a studio, a captioner, or a scheduler.