Give the agent a start frame and a motion — generate_video with image_url turns the picture into video. The agent awaits the result via wait_generation and shows it in the conversation.
The image-to-video scenario uses the same generate_video tool: as soon as the arguments include an image_url with an HTTPS link to a start frame, generation switches from text-to-video to animating the image. This is handy for product shots, portraits, and covers: upload a picture, describe the camera motion — and the agent assembles a short clip without leaving the chat.
| Parameter | Type | Req. | Description |
|---|---|---|---|
image_url | string | yes | HTTPS link to the start frame |
prompt | string | yes | Description of motion and scene |
model | string | — | Slug of a video model with i2v support |
duration | number (сек) | — | Clip duration |
resolution | string | — | 480p / 720p / 1080p |
generate_audio | bool | — | Add audio if the model supports it |
{
"name": "generate_video",
"arguments": {
"image_url": "https://example.com/product.jpg",
"prompt": "gentle parallax, camera slowly pushes in, soft light",
"resolution": "720p",
"duration": 5
}
}{
"request_id": "c4d9...1fa",
"status": "IN_PROGRESS",
"model": "seedance-2-fast-i2v",
"cost_credits": 12,
"next_step": "call wait_generation with this request_id"
}generate_video switches to image-to-video itself when image_url is passed. No separate tool needed.
Animate a product photo or a cover for Reels and Shorts right from the agent chat.
A clipia_test_* key returns a mock instantly — debug the scenario beforehand.
claude mcp add --transport http clipia https://api.clipia.ai/mcp \
--header "Authorization: Bearer clipia_live_XXXX"For claude.ai and ChatGPT no key is needed — sign in with your account. Create a key in Clipia settings.
The credit cost is returned at launch (cost_credits); image-to-video is billed as video generation.
Any public HTTPS link to an image. The agent passes it in image_url.
Same tool, but with image_url — the video builds from your frame, not from scratch.
Same as text-to-video: status is awaited via wait_generation, long-poll up to 30 s per call.