Generate Video
Turn text or stills into video
Try asking
About
Generate Video turns a text prompt or a still image into a video clip through multiple providers behind one MCP call, with resolutions up to 4K, clips from 4 to 30 seconds, and synchronized audio on the models that support it — paid calls start at $0.01 and failed calls are never charged.
Checked 2026-08-26 against this manifest.
Which video model should I choose?
| Model | Clip length | Resolutions | Audio |
|---|---|---|---|
| Seedance 2.5 (default) | 4–30s | 480p, 720p | Yes |
| Google Veo 3.1 | 4, 6 or 8s | 720p, 1080p, 4K | Yes |
| Kling O3 Pro | 5 or 10s | 720p, 1080p | Yes |
| FLUX.3 Video | model default | 720p, 1080p | No |
Call list_models for the live matrix — the catalog changes, and these four are the ones worth starting from. Seedance 2.5 is the default when you omit model.
Why do some calls fail?
Most failed calls are prompts the model refused or aspect ratios it does not support. Failed paid calls are never charged, so a rejected prompt costs nothing; rephrase it, or pick a model whose supported durations and ratios match what you asked for.
What you can do
- text_to_video — generate a video from a text prompt, no source image needed
- image_to_video — animate any still image into a video clip with an optional motion direction prompt
- list_models — browse all available video models with supported durations, resolutions, and audio support
- Choose resolution up to 4K, durations from 4 to 30 seconds depending on the model
- Some models generate synchronized audio alongside the video
Who it's for
Content creators animating product shots or lifestyle imagery, marketers producing short-form social video, and developers building AI video workflows.
How to use it
- First call list_models to compare available options and pick the right model for your duration and quality needs
- Call text_to_video with a scene prompt when you do not have a first frame
- Otherwise generate a first-frame image with the generate-image tool, then call image_to_video with a motion prompt
- Video generation takes time — keep the first attempt short and review the returned video URL
Getting started
Try list_models, then text_to_video from a prompt or image_to_video with a still. Representative manifest-schema shape: video_url_direct, model, duration_seconds [redacted]. Motion accuracy is not guaranteed; timing varies.
Generate a video from a text prompt, with no source image
Animate a still image into a video
Generate a video from a text prompt, with no source image. Check list_models for duration, resolution, and audio support.
Animate a still image into a video. Provide an image URL and optional motion prompt. 9 models with different quality and cost tradeoffs.
List video model capabilities, pricing, durations, resolutions, and audio support. Filter by type.
- Added Sume as a video provider — 48 models including Seedance 2.5, Kling 4, Veo 3.1, Wan 3.0, MiniMax H3, Vidu Q4, HeyGen Video 1 and talking-avatar lip-sync (OmniHuman, Kling Avatar, InfiniTalk). Use a sume- prefixed model key.
- Removed the persona, scene, product, and outfit inputs that had no effect on video output — pass a first-frame image_url or image_urls instead.
- Added Kling O3, FLUX.3 Video, Grok Imagine 1.5, Veo 3.1 Lite, Wan 2.7, and MiniMax Hailuo 3.
- Veo 3.1 Fast on Google now uses the Fast model, not the standard Veo 3.1 ID.
- Google Veo 3 and Veo 2 are retired — use Veo 3.1 instead.
- Higgsfield is now a full video provider — Seedance, Kling, Veo, Sora, Wan, Hailuo, DoP, Genjutsu, and the rest of the official Higgsfield video catalog. Use a higgsfield- prefixed model key or backend.order to route through Higgsfield.
- Added Runway Gen-4.5, Gen-4 Turbo, Aleph 2.0, and Act Two — Runway’s own models only.
- Added MiniMax H3 Max (text, image, multi-angle reference, and lip-sync) and Seedance 2.0 US reference-to-video.
- Higgsfield DoP now requires a still image and accepts camera-motion presets. Popcorn stays an image model.
- Added a model comparison table with clip lengths, resolutions and audio support
- Explained why some generations are refused, and confirmed failed paid calls are never charged
- Text-to-video now works the same way on every client, not just one.
- Video generation is no longer offered through the Claude directory connector, which does not carry AI media generation.
- Text-to-video is available again on the Claude connector. Other clients still animate from a still image.
- Clarified that the public generation skill requires a first frame and uses image_to_video.
- Added request-time same-model provider switching and retryable failover for image-to-video models with multiple provider endpoints.
- Default video model is now Seedance 2.5 — up to 30 seconds with native audio. Seedance 2.0 remains available.
- duration parameter description now lists supported durations per model (Veo 3.1: 4–8s, Kling 3.0: 5/10s, Wan 2.2 Lightning: 5s only, Sora 2: 4/8/12s, etc.) so callers can pick a valid duration without an extra list_models round-trip.
- Removed text_to_video — generate the first frame with generate-image, then animate it via image_to_video for consistent results across all models.
- Added OpenRouter as a video provider — dynamically discovers available video models including Veo 3.1, Sora 2, Seedance, and Wan. Models appear automatically as OpenRouter adds support.
- Accept personas, scenes, products, and outfits from your file library
- Image input now required for all video generation — use generate-image to create the first frame, then pass it to image_to_video. Added Seedance 2.0 and Seedance 2.0 Fast models.
- Style references now supported — pass a style reference when generating videos for consistent visual direction
- Unsupported durations now snap to the nearest valid duration instead of failing
- Added Google as a direct provider — 5 new Veo models: Veo 3.1, Veo 3.1 Fast, Veo 3, Veo 3 Fast, Veo 2. Direct Google API access with per-second billing and 4K resolution support.
- Removed hardcoded model names from descriptions — agents discover models dynamically via list_models
- Added Wan 2.2 Lightning via Prodia — its model registry reports text-to-video and image-to-video capabilities; Generate Video now exposes image_to_video only.
- Added Prodia as an alternative provider — 6 new models (prodia-veo-fast, prodia-veo, prodia-kling, prodia-sora-2, prodia-seedance-pro, prodia-seedance-lite)
- Added subtitle, expanded description, and agent instructions
- Initial release
Powered by
Related
Example workflows
Animate a product image
Turn an existing product image into a short motion clip with a controlled camera or object movement.
- Upload a product image, or create a first frame with Generate Image.
- Ask Generate Video to animate that image with a specific motion prompt.
- Review the clip, then refine its motion, duration, framing, or model if needed.
Guides
- How to create a product demo video with AIAnimate product screenshots into a short clip with Generate Video. Supply frames you own. Generated motion is not a pixel-accurate UI recording.
- How Much Does MCP Cost for a Small Team?Claude and ChatGPT seats are separate. ToolRouter team is $29/seat with shared credits. Free catalog tools stay free. Paid skills start at $0.01.
- How to Turn an Image into Video with AI AssistantsChatGPT, Claude, and Grok do not animate a still by default. Connect ToolRouter, pass an image URL, and ask for the motion. Image in, clip out.
- Can Claude Generate Video? Does Claude Have Sora?Claude has no native Sora or MP4 workflow. Connect ToolRouter, then ask Claude to generate a video clip and return a downloadable file.
Questions
Can it make a video from text alone?
Yes. `text_to_video` generates a clip from a scene prompt, with no source image. `image_to_video` animates a still instead — generate a first frame with Generate Image, then pass it in with an optional motion prompt.
How do I animate a still image?
Pass a ToolRouter file ID or hosted image URL to `image_to_video`, then describe the motion or camera movement you want.
Do the model choices matter?
Yes. Different models trade off quality, speed, cost, and audio support.
What do I get back?
The output is a downloadable video URL.