Short-drama workflow API: from script to clips

POST /v1/workflows/short-drama takes a script and runs these steps in order:

1. Characters: finds the people and animals in the script and describes how they look;
2. Storyboard: splits scenes into shots with framing, camera move, picture, dialogue and sound, then packs shots into clips that fit the video model's length limit;
3. Character sheets: one full-body reference image per character, so they look the same in every shot;
4. First frames: one image per clip, using the characters in it as references;
5. Video clips: each clip animated from its first frame. Dialogue and sound are in the clip prompt, and models with native audio produce sound directly.

The workflow is one async job: submit a script, get a workflow ID, then poll. Each step is an ordinary generation billed at its model's unit price and listed in the console's Logs; failed steps are free.

Prefer no code? See the no-code tools in the AI short-drama solution.

#Estimate the cost

curl https://bigmodel.io/v1/workflows/short-drama \
  -H "Authorization: Bearer $BIGMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"script": "SCENE 1. Convenience store, rainy night…", "max_segments": 6, "dry_run": true}'

{"object": "quote", "workflow": "short_drama", "max_segments": 6, "estimated_cost_usd": 4.24}

The estimate assumes max_segments clips of typical length. You pay for the clips and seconds actually produced, which is usually less.

#Create a workflow

curl https://bigmodel.io/v1/workflows/short-drama \
  -H "Authorization: Bearer $BIGMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: episode-01" \
  -d '{
    "script": "SCENE 1. Corner store, rainy night\nMIA (20, short hair, yellow raincoat) runs under the awning holding a soaked puppy.\nMIA: It is okay, we are home.",
    "style": "warm, cinematic, photoreal",
    "aspect_ratio": "9:16",
    "image_model": "qwen-image-2.1",
    "video_model": "seedance-2.0-fast",
    "max_segments": 6
  }'

import os, time, requests

BASE = "https://bigmodel.io/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['BIGMODEL_API_KEY']}"}

wf = requests.post(f"{BASE}/workflows/short-drama", headers={**HEADERS, "Idempotency-Key": "episode-01"}, json={
    "script": open("episode-01.txt", encoding="utf-8").read(),
    "style": "warm, cinematic, photoreal",
    "aspect_ratio": "9:16",
    "video_model": "seedance-2.0-fast",
    "max_segments": 6,
}).json()

while wf["status"] in ("queued", "running"):
    time.sleep(15)
    wf = requests.get(f"{BASE}/workflows/{wf['id']}", headers=HEADERS).json()
    print(wf["stage"], wf["progress"])

for seg in wf["segments"]:
    print(seg["title"], (seg["video"] or {}).get("url"))
print("total", wf["cost_usd"])

#Request fields

The Idempotency-Key header is optional; repeating a value returns the same workflow.

#Progress and results

GET /v1/workflows/{id} returns:

{
  "id": "e5d2c392-…",
  "object": "workflow.short_drama",
  "status": "running",
  "stage": "videos",
  "progress": {"clips_done": 1, "clips_total": 2},
  "estimated_cost_usd": 0.60,
  "cost_usd": 0.21,
  "characters": [
    {"name": "Mia", "look": "20, short hair, yellow raincoat…", "image": {"task_id": "…", "status": "succeeded", "url": "https://bigmodel.io/v1/media/…"}}
  ],
  "segments": [
    {
      "index": 1, "title": "Shot 01 · wide", "scene": "Corner store, rainy night", "seconds": 4, "cast": ["Mia", "Puppy"],
      "picture": "Warm light spills from the store glass…",
      "video_prompt": "[Shot 1 | 00:00-00:04] …",
      "frame": {"task_id": "…", "status": "succeeded", "url": "https://bigmodel.io/v1/media/…"},
      "video": {"task_id": "…", "status": "running"}
    }
  ],
  "storyboard": {"visualTone": "…", "scenes": [ … ]}
}

One failed clip doesn't fail the workflow: a clip whose first frame failed is skipped and the rest continue. The workflow is failed only if every frame or every clip fails.

#Cancel

curl -X POST https://bigmodel.io/v1/workflows/{id}/cancel -H "Authorization: Bearer $BIGMODEL_API_KEY"

Finished steps stay charged; running generations are cancelled and free.

#Time and cost

#Join the clips into an episode

The workflow returns clips in order. Join them in any editor, or with ffmpeg:

# Download the segment videos in order as 01.mp4, 02.mp4 …
printf "file '%s'\n" *.mp4 > list.txt
ffmpeg -f concat -safe 0 -i list.txt -c copy episode.mp4

#Troubleshooting

#Related

• Multimodal overview
• Video generation
• Image generation
• Media task API reference