Avatar video succeeded but captions failed: what to do on Sume

If your Sume avatar video job succeeded but captions are missing, check captions.status in the result. A failure in the caption stage is a soft failure: the job can still succeed with a clean primary video_url and captions.status set to failed. Take that clean MP4 and run a standalone video caption job on it.

What the docs say

From Generate avatar video, read 2026-10-08:

• Captions are optional and burn into the clean final MP4 after generation.
• They use the spoken script or video_inputs text.
• Sume never adds captions to preview stills.
• Inline captions do not create a separate billed video-caption job.
• For inline captions, Sume rejects an estimated duration of more than 60 seconds.

Check, then recover

Read the result with GET /v1/jobs/{id}/result. If captions.status is failed, you still own a usable uncaptioned clip, so do not rerun the whole avatar job; that would be a second paid generation. Use the standalone route instead:

Mistakes that cause failures

Some failures are avoidable. A Korean script with style slam, punch or tiktok-green is rejected up front with 400 caption_hangul_text_latin_style, because those faces render Hangul as tofu; pick a Hangul style instead. A silent clip has no speech to transcribe, and the standalone job returns caption_no_speech with next_action use_overlay_captions. In that case send authored cues with text, start and end instead.

The standalone job accepts style, font, language and script_text. Sending script_text avoids relying on speech-to-text, which helps with names and product terms. See Video captions.

Preventing caption failures

Send a script whose language matches the caption style. For Korean, choose a Hangul style such as korean-ad, black-outline or pill-karaoke. For English, slam is the default. Keep the estimated duration at or below 60 seconds when using inline captions.

If speech is unclear, use the standalone caption route with script_text so the words come from your text and not from transcription. For brand names, this avoids misspellings.

Keep both versions of a video when you re-caption: the clean MP4 and the captioned one. A clean master lets you try different styles later without paying for a new avatar render.