Approve the micro-drama lead's first frame, then render
Create an Avatar video preview, review the first-frame stills, regenerate if needed, then call generate-video. Quality can change at the final step.
How do you check how a recurring AI character looks before paying for the full video? Create an Avatar video preview: it generates only the first-frame still stage, you review it, and you call generate-video on the preview id once it is right. The final render tier can still be changed at that last call without a new preview.
This matters for micro-dramas because the lead has to look the same in every episode, and a wrong framing found after the render is a full render wasted.
The four calls
The preview resource has four routes: POST /v1/avatar-video-previews to create, GET /v1/avatar-video-previews/:id to read, POST .../regenerate to refresh stills, and POST .../generate-video to start the real render. The create body is the same as Avatar Video: a script or video_inputs, plus optional product_image, scene, quality, aspect_ratio, title and captions.
When the preview is ready, preview_image_url carries the primary still and scene_previews[] carries one still per input scene. For multi-scene previews with a shared scene, later stills are pose-anchored continuations of the first frame, which is what you want for a lead who must not drift between scenes.
What needs a new preview
Regenerate reuses the stored request and refreshes only the first-frame stills. Structural edits cannot be patched in at generate time.
| Change | What to call |
|---|---|
| Another take of the same framing | regenerate (same avatar_preview id) |
| Different quality tier for the final render | generate-video with quality |
| Edit script or video_inputs | new preview |
| Different avatar_handle, scene or aspect_ratio | new preview |
Captions and quality
Caption intent is stored on preview create and applied only at generate-video time; Sume never burns captions into preview stills. An empty body (or {}) keeps the quality you chose at create, and the optional quality field overrides only the final render. Admission, pre-spend, ledger reservation and provider submit all use the effective tier.
The window is the same as direct Avatar Video: an estimated 4-60 seconds inclusive. Media inputs are public HTTPS URLs.
A practical loop
Create the preview while you settle the lead's outfit, background prompt and framing (preview stills do not depend on the quality tier). Regenerate until the still is right. Then call generate-video with quality plus or max for the episode that ships. Read the finished resource with GET /v1/avatar-videos/:id and keep the job id next to the episode number so a later re-render is traceable.
Why the preview step pays
A preview exists so you can reject a face before the expensive render. The first-frame stills show how the lead, framing and any product look, and you can regenerate until they are right. Only then do you call generate-video on the preview id.
Because captions are stored when the preview is created and applied when the video is generated, decide the caption style up front. Changing the quality tier at the last call affects the final render only, not the stills you approved.
- Reject a bad frame at preview time, not after rendering.
- Use regenerate for edits that keep the same preview; start a new preview when the script or avatar changes.
- Record the approved preview id next to the episode.
Cost shape
The final render is billed at the talking-video rate for the chosen tier, and the quality override on generate-video changes only that last step. The docs do not publish a separate preview price here, so read the live number in GET /v1/catalog before budgeting a season.
Sources
Related posts
More in Sume Avatar 1.0
- Avatar 1.0 product image surcharge: $0.01 to $0.03 per second
Adding product_image to an Avatar 1.0 UGC ad adds $0.010, $0.013 or $0.030 per second at standard, plus and max. Thirty 15-second plus ads cost $116.10.
- Avatar handle with an @ prefix: Sume stores it without the @
You can send an avatar handle as @product_host or product_host. Sume normalizes it and stores it without the @. What that means for lookups and clips.
- Avatar holiday greeting for 12 clients: $23.03
One avatar, twelve 10-second personalized greetings at standard quality: $0.95 plus 12 clips at $1.84 each is $23.03 on Sume.
- Avatar video aspect ratio for a Shorts series: 9:16, 1:1 or 3:4?
Avatar 1.0 renders 1:1, 3:4, 9:16, 4:3 and 16:9 (default 9:16) at 720p. YouTube Shorts series accept portrait or square only, so 4:3 and 16:9 are out for them.
Written by Sume