How to make a TikTok micro-drama with AI: six steps and the costs
A six step workflow for a vertical AI micro-drama episode with Sume: script, shots, first frames, assembly, captions and the is_aigc flag. About $52 for 90 s.

Make a 90 second vertical micro-drama episode by writing three 30 second shots, generating them with Seedance 2.5 at 9:16, joining them with Timeline 1.0 and setting is_aigc true when you post. The generation costs about $52.00 and the join $0.20. Micro-dramas were reported as gaining traction on TikTok by Metricool.
Below are the six steps in the order you do them, with the Sume limit that matters at each one.
The six steps
Each step has one decision and one number. Retries are new jobs, so budget for one in five shots being redone.
| Step | What you do | Sume number |
|---|---|---|
| 1 Script | Three 30 s shots, ending on a hook | Clips run 4 to 30 s |
| 2 First frames | Make stills of the lead and set | Up to 16 reference images on gpt-image-2.5 |
| 3 Shots | Generate with frame_images and 9:16 | $0.5778 per second at 720p |
| 4 Assemble | Join shots in order | $0.10 per started minute |
| 5 Captions | Burn dialogue into the cut | $0.20 per job up to 60 s |
| 6 Post | Set the AI flag | is_aigc on TikTok |
Steps 2 and 3: keep the lead the same
Generate a still of your lead first and pass it as frame_images with frame_type first_frame. For the next shot, grab the last frame of the previous one with video frames and use it as the next first frame, so the scene flows. See chaining clips from the last frame.
Steps 4 and 5: assemble and caption
Timeline renders up to 200 video slots and defaults to 1080x1920. Fades up to one second are allowed between slots. If you caption each 30 second shot before assembly instead of the whole episode, you pay $0.20 per shot, so $0.60 per episode, against $0.20 for one pass over the finished cut when it is 60 seconds or less. A 90 second cut exceeds the 60 second caption limit, so caption in two passes or per shot.
Step 6: mark it as AI
TikTok's Content Posting API has an is_aigc field. The reference says to set it true if the video is AI generated, and the video is then labelled with the "Creator labeled as AI-generated" tag in its description. The same reference lists video_cover_timestamp_ms for the cover frame and disable_comment. Read it at TikTok for Developers.
What Sume does not do
Sume does not write the plot, post to TikTok or guarantee the same face across unrelated prompts; consistency comes from your reference images. A regenerated shot is a new generation and is not promised to match the first.
Sources
Related posts
More in Use cases
- How to remix a video with AI: four routes by what you change
Remixing a clip means a prompt edit, a person swap, a motion transfer, or a new clip from references. Which Sume route fits each, with the fields.
- Ideogram 4 tops out at 2048 px: upscale to 4K with Sume
Ideogram 4 outputs run 256 to 2048 px per the Hugging Face card. To go bigger, send the file to Sume's image upscale endpoint: public URL, factor, poll.
- Incompetech CC BY 4.0 credit line: wording and where to put it
Kevin MacLeod's Incompetech FAQ gives the exact credit wording, where it must sit, and when a TV or radio spot needs a Standard License instead.
- AI info on the post or in the menu? Generated vs edited clips
Meta moved AI info for edited content into the post menu while generated content keeps the visible label. Which Sume job types fall on which side.
Written by Sume