Talking-head ad: Omni audio, Kling audio on, or Avatar Video?

For a talking-head ad, use Avatar Video when you need the same face over 15 to 60 seconds, gemini-omni-flash-1.1 when you want a single scripted scene of 3 to 10 seconds with synced audio, and kling-3 with audio on when you need 4 to 15 seconds and a 1:1 frame. Avatar Video takes a ready avatar and a script on POST /v1/avatar-1.0/talking-video; the two video models take a prompt on POST /v1/videos.

Everything below is from Sume's Video Generation and Generate avatar video pages, read 2026-10-05.

Three routes side by side

Sume bills the provider list price times 1.25 and rounds the job up to the cent, so a one-second figure here is a rate, not a charge.

How to choose

The deciding input is whether the face must repeat. Avatar Video starts from avatar_handle, so the same presenter appears in every clip, and it accepts quality of standard, plus (the default) or max. A video model starts from a prompt, so each take can show a different person unless you pin one with a first-frame photo or, on Omni, a reference image.

Length is the second filter. A 10-second Omni clip at 720p is $1.25. A 15-second Kling clip with audio is $3.15. A 30-second spoken ad does not fit either model in one take, so use Avatar Video or two clips joined on a Timeline.

• One face, many scripts: Avatar Video.
• One 3 to 10 s scripted scene with room sound: Omni.
• A 1:1 talking clip of 11 to 15 s: Kling 3 with audio on.
• Silence beats and scene changes inside one video: Avatar Video video_inputs.

Not on the list

Veo 3.1, Luma Ray3.2, Runway's own models and HappyHorse have no Sume model id, so a brief written for one of them has to be moved to one of the routes above. Check GET /v1/videos/models before you plan around a new name.