| name | create-video-seedance-2-0 |
| description | Create AI videos with KIE.AI Bytedance Seedance 2.0 Fast. Use when the user wants text-to-video, image-to-video with first/last frames, multimodal reference-to-video, optional generated audio, polling task status, or downloading Seedance results through the KIE Market API. |
Create Video Seedance 2.0
Rules
- Use this skill for KIE.AI
bytedance/seedance-2-fast video generation.
- Read
.env from the repo root before running scripts. Require KIE_API_KEY; never print the key.
- If inputs are local files, upload them to KIE file storage first. Uploaded files are temporary and expire after 3 days.
- Generated result URLs are temporary; download finished videos into the job folder immediately.
- Do not use
callBackUrl in this repo. The local environment has no public callback receiver, so every task must be checked with a polling loop through recordInfo.
- For Vietnamese user-facing output, write Vietnamese with accents. Keep API fields, model IDs, paths, and enum values unchanged.
Model
KIE Market endpoint:
POST https://api.kie.ai/api/v1/jobs/createTask
GET https://api.kie.ai/api/v1/jobs/recordInfo?taskId=<taskId>
Authorization: Bearer <KIE_API_KEY>
Create body:
{
"model": "bytedance/seedance-2-fast",
"input": {
"prompt": "A cinematic vertical product shot...",
"first_frame_url": "https://...",
"last_frame_url": "https://...",
"reference_image_urls": ["https://..."],
"reference_video_urls": ["https://..."],
"reference_audio_urls": ["https://..."],
"return_last_frame": false,
"generate_audio": false,
"resolution": "720p",
"aspect_ratio": "16:9",
"duration": 15,
"web_search": false
}
}
Scenario Selection
Choose exactly one generation mode:
- Text-to-video: prompt only, no media URLs.
- Image-to-video first frame:
first_frame_url only.
- Image-to-video first and last frames:
first_frame_url + last_frame_url.
- Multimodal reference-to-video: one or more
reference_image_urls, reference_video_urls, or reference_audio_urls.
Do not mix first/last frame fields with reference_*_urls. KIE notes these scenarios are mutually exclusive. If strict first/last frame identity is required, prefer first_frame_url + last_frame_url over multimodal references.
Recommended Defaults
- TikTok/Reels/Shorts:
--aspect-ratio 9:16 --resolution 720p.
- General horizontal video:
--aspect-ratio 16:9 --resolution 720p.
- Keep
--generate-audio off unless the user asks for model-generated sound.
- Keep
--web-search off unless the user asks for current/real-world grounding.
- The
generate command always loops on GET /api/v1/jobs/recordInfo?taskId=... until success or fail.
Script
Generate and poll in one command:
python skills/create-video-seedance-2-0/scripts/create_video_seedance_2_0.py generate \
--env-file .env \
--prompt "A cinematic vertical shot of a Vietnamese street food stall at night, handheld camera, warm practical lights" \
--aspect-ratio 9:16 \
--duration 15 \
--output-dir jobs/<job_id>/input/raw_assets/videos/seedance
First/last-frame image-to-video:
python skills/create-video-seedance-2-0/scripts/create_video_seedance_2_0.py generate \
--env-file .env \
--prompt "Animate the subject with a slow dolly-in, natural cloth movement, realistic lighting" \
--first-frame jobs/<job_id>/input/raw_assets/images/start.png \
--last-frame jobs/<job_id>/input/raw_assets/images/end.png \
--aspect-ratio 9:16 \
--output-dir jobs/<job_id>/input/raw_assets/videos/seedance
Multimodal reference-to-video:
python skills/create-video-seedance-2-0/scripts/create_video_seedance_2_0.py generate \
--env-file .env \
--prompt "Create a polished product lifestyle clip using the reference product and the audio rhythm" \
--reference-images jobs/<job_id>/input/raw_assets/images/product.jpg \
--reference-audios jobs/<job_id>/input/raw_assets/audio/beat.mp3 \
--aspect-ratio 9:16 \
--output-dir jobs/<job_id>/input/raw_assets/videos/seedance
Check a task later:
python skills/create-video-seedance-2-0/scripts/create_video_seedance_2_0.py status \
--env-file .env \
--task-id task_bytedance_1765186743319 \
--output-dir jobs/<job_id>/input/raw_assets/videos/seedance \
--download
Outputs
The script writes:
<output-dir>/seedance_task_<taskId>.json
<output-dir>/seedance_result_<taskId>_<n>.<ext>
Use generated videos as raw assets. After download, continue with the usual pipeline:
.venv/bin/python -m tools.asset_index.exporter jobs/<job_id>/input/raw_assets/ --output jobs/<job_id>/source/asset_semantics.toml
Troubleshooting
401: missing or invalid KIE_API_KEY.400 / 422: invalid parameter combination; re-check scenario exclusivity.fail state: read failCode and failMsg from the saved task JSON.- No result file downloaded: inspect
resultJson; if URLs are nested differently, download the returned result URL manually and update the script if this becomes a recurring shape.
Sources
