90-second TikTok mini-drama episode: avatar videos cap at 60 seconds

A single Sume avatar talking-video job covers 4 to 60 seconds, so a 90-second mini-drama episode is two jobs of about 45 seconds each, joined into one MP4. Timeline 1.0 bills $0.10 per output minute rounded up, so a 90-second render is 2 minutes, or $0.20.

TikTok's own help page on Growth Max mini-drama campaigns does not state an episode length, so the 90 seconds here is your creative choice, not a TikTok requirement.

Split the script at a scene break

Generate avatar video accepts a script that estimates to 4-60 seconds per job and defaults to 9:16 at 720p. Cut at a natural beat: a door closing, a reveal, a pause. Cutting mid-sentence gives you a visible jump.

The same page describes multi-scene video_inputs with silence beats inside one job. Use them for pauses within a scene; use a second job for a new scene, setting, or avatar.

Join the parts

Timeline 1.0 builds the final file from an audio spine, which can be 1-1800 seconds with up to 20 audio.parts. Run the plan call first; it is unbilled and shows what the render would do. Output defaults to 1080x1920, and output.fade_in_seconds and output.fade_out_seconds accept 0-5 seconds.

Add music last, with duck_db so it drops under dialogue. The ducking post covers the settings.

Keep the avatar consistent

Use the same avatar handle for both jobs, so the face and voice do not drift between halves. The avatar docs describe the handle you get from creating an avatar.

Budget and review before posting

Run the Timeline plan call before the paid render. It is unbilled, so you can change the join points, the fades, and the audio parts as often as you like and pay only for the final render. Each render is billed per output minute rounded up, which means a 61-second cut and a 120-second cut both bill as two minutes. If you can trim an episode to 60 seconds exactly, the same join costs $0.10.

After the join, check the seam. Look at the last frame of scene A and the first frame of scene B: the avatar's posture, lighting, and background should match. If they do not, adjust the script break so the cut falls on a pause, or add a short silence beat at the end of scene A. Then check the audio with a quick listen at the cut, since a breath or a room-tone change is more noticeable than a picture jump.