| name | coze-video-gen |
| description | Generate a video from text prompts and/or image inputs using Coze video generation. Use when you need text-to-video, image-to-video, first-frame control, last-frame control, or a returned last-frame image URL. |
| homepage | https://www.coze.com |
| metadata | {"openclaw":{"emoji":"🎬","requires":{"bins":["node"],"config":["plugins.entries.coze-openclaw-plugin.config.apiKey"]}}} |
Coze Video Generation
Generate a video URL from text and image inputs using Coze.
This skill runs as a Node CLI wrapper around the backend SDK. Do not use the SDK from client-side code.
Quick start
node {skillDir}/scripts/gen.mjs --prompt "A serene mountain landscape with flowing clouds"
node {skillDir}/scripts/gen.mjs --prompt "A cinematic robot walking through neon rain" --duration 6 --ratio 9:16 --resolution 1080p
node {skillDir}/scripts/gen.mjs --prompt "Animate this concept art" --first-frame "https://example.com/first.png"
node {skillDir}/scripts/gen.mjs --prompt "Transition from sunrise to night" --first-frame "https://example.com/start.png" --last-frame "https://example.com/end.png" --return-last-frame true
Options
--prompt <text> text prompt
--image <url> reference image URL; repeat the flag to pass multiple images
--first-frame <url> first frame image URL
--last-frame <url> last frame image URL
--model <id> model id, default SDK model is doubao-seedance-1-5-pro-251215
--duration <seconds> video duration in seconds
--ratio <ratio> one of 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, adaptive
--resolution <value> one of 480p, 720p, 1080p
--watermark <true|false> whether to keep watermark
--seed <n> fixed random seed for reproducibility
--camera-fixed <true|false> whether to fix the camera
--generate-audio <true|false> whether to generate synced audio
--return-last-frame <true|false> whether to return the last-frame image URL
--max-wait-time <seconds> max synchronous wait time
--callback-url <url> async callback URL passed through to the SDK
--header <key:value> custom HTTP header; repeatable, alias -H
Behavior
- At least one of
--prompt, --image, --first-frame, or --last-frame is required.
- Successful runs print
Task ID, Status, Video URL, and optionally Last Frame URL.
- The CLI does not download video files locally.
- Printed video URLs and frame URLs must be kept exactly intact, complete, and accurate. All URL parameters must be preserved without truncation, rewriting, omission, or reordering; in particular, parameters inside the query string such as
sign must not be dropped, otherwise the asset may be inaccessible.
- Unless the user explicitly asks to download the URL content, only return the complete URL link to the user.
Notes
- The skill runtime requires
plugins.entries.coze-openclaw-plugin.config.apiKey.
{skillDir} means the directory containing this SKILL.md.
- The default model
doubao-seedance-1-5-pro-251215 supports text input plus first_frame and last_frame. Generic reference_image inputs from --image may be rejected by that model.
- Only use
--image when you have selected a model that explicitly supports reference_image.
- The returned video URL is already hosted with a valid expiration period. Unless the user explicitly needs rehosting or download, use it directly.