Same set every episode: an Avatar 1.0 scene photo or prompt background

Avatar Video 1.0 scenes are a prompt or a photo. Use one photo for the set in every episode, or reuse one prompt word for word, to keep the same place.

4 min readSume
All posts

Use the same photo scene in every episode. In Avatar Video 1.0 a scene is either {type: prompt} or {type: photo, image_url}, and the photo form puts the presenter in the place your image shows. Keep that image at one public HTTPS URL in your season config so every episode starts from the same set.

The two scene types

Scene types for Avatar Video 1.0 (read 2026-10-07)
SceneFieldBest for
Prompt{type: "prompt"} with a text descriptionExploring a look, a one-off episode
Photo{type: "photo", image_url}, a public HTTPS imageA recurring set, a real office or studio
Silence beatvoice.type: "silence" with a required duration, no scriptA pause between lines in multi-scene videos

Why a photo wins for a series

A prompt is interpreted again every time, so the room varies a little from one episode to the next. A photo is a fixed input. If your set is a real desk, take one photo of it, keep the lighting the same on shoot days, and use that file for every episode. Change the photo on purpose when the story moves somewhere else.

Keep it stable

  • Host the image at a durable HTTPS address, not a temporary link. Sume rejects localhost and private-network URLs.
  • Leave space in the photo where the presenter will appear; a cluttered centre fights with them.
  • Only one resolved avatar is supported per final video, so a two-presenter episode needs two videos cut together on a timeline.
  • Preview first with POST /v1/avatar-video-previews and regenerate before you pay for the full video.

When to use a prompt

A prompt scene is quicker when the set changes every episode, such as a travel series. Write the scene once as a sentence, hold it in your season config and reuse the exact words. Small wording changes give a different room, so edit the text only on purpose.

Both types work in multi-scene video_inputs, so you can open an episode on a photo of the set and cut to a prompt scene for a flashback, as long as the same avatar appears throughout.

Write down, once, which scene type each recurring location uses and keep that table in your repo. In practice a series has two or three settings, such as an office, a kitchen and a street, and each one maps to a single photo URL or a single sentence. When an editor asks why episode 8 looks different, the answer is then one line in the config.

Check one more thing before a season: the photo's orientation should match the aspect ratio you request. A landscape photo in a 9:16 video crops, and what survives the crop is a choice you want to preview, not discover.

Sources

Related posts

More in Sume Avatar 1.0

All Sume Avatar 1.0 posts

Written by Sume