Generate avatar video
Avatar videos turn a ready avatar into a script-driven talking video. Prefer the canonical Avatar 1.0 route:
Launch requests reference a ready avatar with top-level avatar_handle (or
per-scene character fields inside video_inputs). Provide exactly one of
script or video_inputs.
Scripts and multi-scene plans are accepted when Sume estimates the target video duration at 4-60 seconds inclusive. Shorten longer scripts or split them into multiple jobs.
Create an avatar video
Create an avatar talking video
POST /v1/avatar-1.0/talking-video
Required
Optional product and scene inputs
- Omit
product_imagefor a productless avatar video. - Use
scene: { "type": "prompt", "prompt": "..." }for scene direction. - Use
scene: { "type": "photo", "image_url": "https://..." }for a photo scene reference. - Media fields must be fetchable public HTTPS URLs. See Media inputs.
Quality
Avatar Video accepts quality: "standard" | "plus" | "max".
| Value | Behavior |
|---|---|
plus | Default when omitted. Balanced quality path. |
standard | Fastest Sume execution path. |
max | Highest quality tier; slower turnaround. |
Aspect ratio and resolution
aspect_ratiosupports1:1,3:4,9:16,4:3, and16:9. Default:9:16.resolutionis currently720p.
video_inputs
Use ordered video_inputs instead of a single script when you need scene
hooks, demos, or silence beats in one composed video. Total planned duration
must still land in the 4-60 second window.
voice.type: "silence" is a non-speaking beat: duration is required, and
script / input_text are not allowed. Spoken scenes still use
type: "text" with exactly one of script or input_text.
Current execution supports one resolved avatar per final video and expects scene backgrounds to resolve to one shared scene.
Inline captions
Optional captions burns styles into the clean final MP4 after generation,
using the spoken script / video_inputs text. Preview stills are never
captioned.
captionstakes the same four knobs as standalone Video captions:style, optionalfont, alanguagehint andscript_text.- Styles:
slam(default),punch,tiktok-green,korean-ad(Hangul karaoke for Korean speech), plus the Hangul identitiesweight-shift,black-outline,highlight,pill-karaoke,clip-wipeandeditorial-emphasis. - A Korean script with
style: "slam"(orpunch/tiktok-green) is rejected with400 caption_hangul_text_latin_stylerather than re-styled — those faces render Hangul as tofu. Pick a Hangul style for Korean speech. - Estimated duration above 60 seconds is rejected for inline captions.
- Caption stage failures soft-fail: the avatar job can still succeed with a
clean primary
video_urlandcaptions.status=failed. - Inline captions do not create a separate billed video-caption job. For captioning an existing public video URL, use Video captions.
Poll and recover
Completed results can include public media.sume.com video artifacts plus
public-safe preview fields such as preview_image_url and scene_previews.
Read avatar-video resources
Preview first, then generate
To review first-frame stills before paying for a full render, create an
Avatar video preview, then call
generate-video on the preview id.
Compatibility aliases
| Alias | Notes |
|---|---|
POST /v1/models/sume/avatar-1.0/talking-video/runs | Canonical model-run alias. Prefer /v1/avatar-1.0/talking-video for new integrations. |
POST /v1/models/sume/avatar-video/v1.0/runs | Legacy launch alias. Same body contract. |