Generate Video
generate_videoGenerate Switch video across the real provider lineup (Kling, BytePlus Seedance, Switch Video/WAN 2.7, Switch Video Edit, Topaz upscale) and modes (text-to-video, image-to-video, frame-to-frame, motion, omni, reference-to-video, video-edit, upscale). ALWAYS call load_video_workflow first and list_video_models second. Use Switch MCP directly; never open the website, browser, or desktop unless the user explicitly asked. For realistic reference-driven video, evaluate BytePlus Seedance 2.5 first unless the user chose another provider or model. Before submission, map every actual image, video, and audio chip to its role and preserve its thumbnail, type, order, and role exactly; never silently drop, merge, reorder, convert, or repurpose references. Pass one shot, or shots:[...] for a storyboard (max 4 by default, hard max 10) where EACH shot is DIFFERENT — never repeat one prompt to get copies. Renders async (~30-90s); a background job delivers each clip to your library. Returns a task_id per shot — poll get_video_status or list_my_videos.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| mode | No | Video mode. Must be supported by the chosen model (see list_video_models). | |
| audio | No | Generate audio. ON by default on Seedance 2.5 (text, image and reference), on Omni, and on every WAN 3.0 lane (text, image, reference); set false for a silent clip. Models without audio ignore this. See list_video_models for which models generate audio and the max seconds with vs without audio. | |
| model | No | Model id from list_video_models (e.g. kling-v3, seedance-2.0-t2v, wan-3.0-t2v, h3-max-t2v, topaz). Or prefer option_id from list_video_models. MiniMax H3 Max ids: h3-max-t2v, h3-max-i2v, h3-max-r2v (option ids h3max-text / h3max-image / h3max-reference). MiniMax H3 Max Turbo (preview, about twice as fast at half the price, no reference lane): h3-max-turbo-t2v, h3-max-turbo-i2v (option ids h3max-turbo-text / h3max-turbo-image). WAN 3.0 ids: wan-3.0-t2v, wan-3.0-i2v, wan-3.0-r2v (option ids wan30-text / wan30-image / wan30-reference): 720p or 1080p, any whole second 2 to 30, generated audio on by default, image lane takes image_url and an optional end_image_url, reference lane takes up to 10 reference_image_urls and 5 reference_video_urls. | |
| shots | No | A storyboard of 1-10 DISTINCT shots. Each item takes the same fields as a single shot (subject, model, mode, image_url, etc.). | |
| subject | No | The shot: subject + motion + scene (video needs motion language, e.g. "slow push-in"). WAN 3.0: refer to references by their place ("Image 1", "Video 1"), not by @tags; there is no negative-prompt field, so write exclusions inline as "Avoid: ...". MiniMax H3 Max: put spoken words in double quotes and the character says them lip-synced; name references by place (Image 1, Video 1, Audio 1). | |
| duration | No | Clip length in seconds, or "auto" to let the model choose. Seedance 2.5 does 4-30s (auto bills the 30s cap up front and refunds the unused seconds); Seedance 2.0 does 4-15s; Switch Video (WAN 2.7) does 5/10/15; WAN 3.0 takes any whole second from 2 to 30; Kling/Switch Video Edit cap at 10; MiniMax H3 Max takes whole seconds 5 to 15 (no auto) — see each model's durations in list_video_models. | |
| image_url | No | Required for image-to-video / frame-to-frame / motion. Accepts EITHER a Switch asset id (from show_media / list_my_assets / upload_media) OR a public https url. An asset id is resolved server-side, so just pass the id you have — no need to fetch a url first. | |
| option_id | No | Optional catalog id from list_video_models (e.g. "kling-image"); use instead of model+mode. | |
| task_type | No | Seedance 2.5 reference mode only: declare what you are doing with the reference clip so the size and length rules are checked immediately instead of failing a minute in. auto = a new take from the references (default), edit = change something inside the clip, extend = continue the clip. Editing and extension inherit the source clip's size, and editing also inherits its length. | |
| video_url | No | Required for video-edit and upscale (the source clip). Accepts one of YOUR Switch videos — a job id from list_my_videos / get_video_status, or its download_url / view_url — or any publicly downloadable https URL. Switch resolves its own videos for you; no need to scrape a page for the file. | |
| resolution | No | Output resolution. Defaults to 1080p where the model supports it. 720p is cheaper and faster. 480p is the cheapest. Seedance 2.5 offers 480p, 720p and 1080p on every mode (no 2K/4K). 4K is only on Kling v3 text/image and Kling Omni; Seedance text-to-video is 720p only. MiniMax H3 Max offers 480p and 768p only (its reference lane is 768p only). Each model lists its available resolutions in list_video_models. | |
| aspect_ratio | No | e.g. 9:16, 16:9, 1:1. Must be allowed for the model (see list_video_models). | |
| end_image_url | No | End frame for frame-to-frame mode. | |
| face_reference_ids | No | Face reference asset ids from upload_reference_asset (frame_type "face") — the ONLY way to use a face/likeness reference in video. Each id is verified server-side (your own untouched original + identity verification) before the shot fires or is charged; URLs and generic uploads here are rejected. | |
| reference_audio_urls | No | Seedance reference/omni only: up to 3 reference audio files to drive synthesized audio. Requires at least one reference image or video. MiniMax H3 Max reference (h3max-reference) also takes up to 3 audio files as conditioning (voice/sound guidance); it generates native audio on every clip regardless. | |
| reference_image_urls | No | GENERIC reference images (products, scenery, outfits, style). Each entry accepts EITHER a Switch asset id (from show_media / list_my_assets / upload_media / get_my_active_references) OR a public https url — asset ids are resolved server-side. Seedance reference/omni accepts up to 9; Kling Omni up to 7. WAN 3.0 reference (option wan30-reference, model wan-3.0-r2v) accepts up to 10; name each one by its place in the prompt ("the woman in Image 1", "the jacket from Image 3"). For Seedance, at least one image or video reference is required. For a person's face/likeness use face_reference_ids instead. | |
| reference_video_urls | No | Seedance reference/omni and WAN 3.0 reference. WAN 3.0 (wan30-reference): up to 5 clips, 15 seconds combined, each at least 16 fps, addressed in the prompt as "Video 1", "Video 2"; the audio inside the clips is not kept. Seedance: reference video clips for motion/style guidance — up to 10 on Seedance 2.5 (clips 2-30s, 30s combined), up to 3 on Seedance 2.0 (see list_video_models for each model's caps). A Seedance video ref can satisfy the required visual anchor. Per the BytePlus guide a video reference is a subject reference: it can carry appearance, identity, motion AND voice timbre. State in the prompt what each asset provides. MiniMax H3 Max reference (h3max-reference): up to 3 clips, each second of reference clip is billed on top of the output seconds (see its notes). | |
| character_orientation | No | Motion mode only: follow the character image (default) or the reference video. | |
| person_rights_confirmed | No | Required with video_people_declaration "person": confirms you own or are authorized to use the person's likeness in the reference video. | |
| video_people_declaration | No | Required with reference_video_urls: "none" confirms no real person appears; "person" runs the protected pipeline (verified account, your own stored upload, Seedance 2.5/Mini lane, provider asset registration) — also pass person_rights_confirmed: true. |