Avatar preview: regenerate stills or create a new preview
Regenerate redraws the first-frame stills of one preview. A new script, avatar, scene or aspect ratio needs a new preview. Quality is set at generate-video.
Use regenerate when the composition is almost right and you only want new first-frame stills from the same stored request. Create a new preview when you change the script, video_inputs, avatar_handle, scene or aspect_ratio, because those structural fields cannot be edited on an existing preview.
Quality is the odd one out. Preview stills are tier-independent and always reused, so you pick or change quality at generate-video, not by regenerating.
Which change goes where
Regenerate reuses the stored preview request (avatar, script or video_inputs, scene, quality, aspect ratio) and refreshes only the stills. It returns the same avatar_video_preview_id with a new preview-only job, so you poll that job like any other. The result is a new draw of the same brief, which suits a first frame with an odd crop or an unlucky expression.
| What you want to change | Do this |
|---|---|
| A different draw of the same first frame | POST /v1/avatar-video-previews/:id/regenerate |
| Script or video_inputs | Create a new preview |
| Avatar (avatar_handle) | Create a new preview |
| Scene | Create a new preview |
| Aspect ratio | Create a new preview |
| Final render quality (standard, plus or max) | Send quality on generate-video |
Test hook lines at scale without paying for full renders
For a UGC-style test with several hook lines, each hook is a different script, so each one is its own preview. That is the cheap part: a preview generates only the first-frame still stage and does not start the full talking-video render. Review the stills, regenerate the ones with bad framing, and only call generate-video on the winners.
The duration window is the same as for Avatar Video: an estimated 4 to 60 seconds inclusive. Media inputs such as product_image and scene.image_url must be public HTTPS URLs.
Generate the winner
An empty body, or {}, keeps the quality chosen at preview create. Sending quality overrides only the final render tier, and admission, pre-spend, ledger reservation and provider submit all use the effective tier. Captions stored on the preview are applied at this step; preview stills are never burned with captions.
Poll the returned job, then read the resource from /v1/avatar-videos/{id}. Prefer resource_status and job_status over the legacy status field when you check readiness.
curl -X POST https://api.sume.com/v1/avatar-video-previews/avp_123/generate-video \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "quality": "max" }'
Sources
Related posts
More in Sume Avatar 1.0
- Avatar video previews: test three holiday hook lines, render only one
Create three Sume avatar-video previews with different hooks, compare first frames, then call generate-video on the winner and keep its stills.
- Avatar video price by length: 4, 15, 30, 45 and 60 seconds
What a Sume Avatar Video 1.0 clip costs at 4, 15, 30, 45 and 60 seconds on standard, plus and max, with and without a product image. Table from the rate card.
- Sume Avatar Video quality default: what plus costs vs standard
Omit quality and Avatar Video runs on plus, not standard. On 20-second clips that is $1.22 more each, or $122 over 100 clips. The arithmetic and a safe default.
- Azure photo avatar is 512x512; Sume avatar video is 720p
Azure's photo avatar renders head-only at 512x512 and 25 fps. Sume's avatar video offers 720p in five aspect ratios, from a prompt, traits or a photo.
Written by Sume