AI micro-drama episode cost: 12 lines, two voices, $9.92 to $18.32

Micro-dramas are rising across TikTok, Reels and Shorts, HeyOrca reports (read 2026-10-07). The format is dialogue: short scenes where faces talk. On Sume that is the one case where the model choice is already settled by the docs. A shot where a person speaks on camera is Fabric with an accepted still plus TTS; video models do not lip-sync to generated speech. So the episode is a stack of TTS lines, a stack of Fabric clips, and one render.

This page prices a concrete episode: 12 lines, two characters, 8 seconds of audio per line, which is 96 seconds in total. Stills for the two characters are not counted here.

Bill of materials

How the pieces connect

• Run each line through POST /v1/tts-1.0/generate with a voice.id per character; a TTS job returns an audio_url, duration_seconds and word timings.
• Send that URL and the measured duration to Fabric (POST /v1/veed/fabric-1.0) with the character's still. Fabric accepts up to 300 seconds of Sume-hosted audio under 10 MB, so an 8-second line is far inside both limits.
• Stitch the 12 clips in one Timeline 1.0 render. If you want the voices as one file first, concat them with timeline audio at $0.01 per job.
• Keep each TTS language equal to the voice language; Sume returns tts_voice_language_warning on a mismatch and asks you to confirm before retrying.

Where the money goes

Speech is a rounding error: the whole script costs 12 cents. Video seconds are the budget, and the lever is resolution. At 480p the episode is $9.92; at 720p it is $18.32. A common plan is to cut the whole episode at 480p, choose the lines that work, and re-run only those at 720p. Re-running 4 of 12 lines at 720p adds 4 x 8 x $0.1875 = $6.00 to the draft.

Cut dead air in the script, not in the render: every second of audio you send to Fabric is billed, including pauses.