5. Video Generation

Video Generation creates video clips from text prompts and optional references. Use it when you want generated footage or reference-guided video changes from the canvas.

How It Works

Describe the shot, subject, style, camera motion, and desired duration. Mosaic sends the prompt and selected compatible references to the generation model, then returns a video asset that can feed downstream tiles. Use Video Generation for:

Product shots and cinematic b-roll
Social clips from a creative prompt
Reference-guided motion from uploaded images or clips
Sound-enabled generated scenes

MiniMax H3 generates 480p, 768p, 2K, or 4K clips of 5 to 15 seconds with native audio. Without selected context it runs text-to-video; with any selected image, video, or audio it runs reference-to-video, which accepts up to 9 images, 3 videos, and 3 audio clips (12 files total). Wan 3.0 generates auto or 2 to 30 second clips with native audio at 480p, 720p, or 1080p; auto lets the model pick the length from the prompt and reference media. Without selected context it runs text-to-video; with any selected image, video, or audio it runs reference-to-video, which accepts up to 10 images, 5 videos, and 5 audio clips (20 files total). Wan 3.0 costs 2.5 credits per output second at 480p, 5 credits per second at 720p, and 10 credits per second at 1080p. Seedance 2 supports selected image, video, and audio references. Seedance 2.5 generates up to 30 seconds at 480p, 720p, or 1080p and accepts up to 30 images, 10 videos, and 10 audio clips (50 files total); with no references it runs text-to-video, with one image it runs image-to-video, and with anything else it runs reference-to-video. Gemini Omni Flash generates 3 to 10 second clips at 360p, 720p, 1080p, or 4k and supports prompt-only generation, selected image references, or one selected video reference. For Gemini Omni Flash, use either selected images or one selected video, not both. Selected video input uses the first 10 seconds and does not include audio. Gemini Omni Flash costs 1.5 credits per output second at 360p, 5 credits per second at 720p, 7.5 credits per second at 1080p, and 15 credits per second at 4k.

API Usage

Create an agent shell:

curl -X POST "https://api.mosaic.so/agent/create" \
  -H "Authorization: Bearer mk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Seedance product shot",
    "visibility": "private"
  }'

Add a Video Generation node:

curl -X POST "https://api.mosaic.so/agent/AGENT_ID/update" \
  -H "Authorization: Bearer mk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "operations": [
      {
        "op": "create_node",
        "node_type_id": "a2eb3ec2-da7a-4f97-94d2-a9a537548522",
        "params_used": {
          "prompt": "A cinematic handheld shot of a matte black water bottle on a mountain trail at sunrise",
          "model": "seedance-2",
          "resolution": "1080p",
          "duration": "8",
          "aspect_ratio": "9:16",
          "generate_audio": true
        }
      }
    ]
  }'

Then run the agent:

curl -X POST "https://api.mosaic.so/agent/AGENT_ID/run" \
  -H "Authorization: Bearer mk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "callback_url": "https://your-webhook.com/mosaic"
  }'

API Info

Node Params & API Details

Node ID: a2eb3ec2-da7a-4f97-94d2-a9a537548522

Node params

Param	Type	Required	Default	Notes
`prompt`	`string`	Yes	`""`	Generation prompt.
`model`	`"seedance-2" \| "seedance-2.5" \| "gemini-omni-flash" \| "minimax-h3" \| "wan-3.0"`	No	`"seedance-2.5"`	Video model.
`resolution`	`"1080p" \| "4k"` for Seedance 2; `"480p" \| "720p" \| "1080p"` for Seedance 2.5 and Wan 3.0; `"480p" \| "768p" \| "2K" \| "4K"` for MiniMax H3; `"360p" \| "720p" \| "1080p" \| "4k"` for Gemini Omni Flash	No	`"1080p"` for Seedance 2; `"720p"` for Seedance 2.5, Wan 3.0, and Gemini Omni Flash; `"2K"` for MiniMax H3	Output resolution. 4K is billed at 2x normal Seedance. Seedance 2.5 1080p costs more per second than 720p. Wan 3.0 costs 2.5, 5, or 10 credits per output second at 480p, 720p, or 1080p. MiniMax H3 costs 5, 6, 13, or 16 credits per output second at 480p, 768p, 2K, or 4K. Gemini Omni Flash costs 1.5, 5, 7.5, or 15 credits per output second at 360p, 720p, 1080p, or 4k.
`duration`	`"auto" \| "4" ... "15"` for Seedance 2; `"auto" \| "4" ... "30"` for Seedance 2.5; `"3" ... "10"` for Gemini Omni Flash; `"auto" \| "2" ... "30"` for Wan 3.0	No	`"auto"` for Seedance 2 and Seedance 2.5; `"8"` for Gemini Omni Flash; `"5"` for Wan 3.0	Duration in seconds. Wan 3.0 `"auto"` lets the model pick the length from the prompt and reference media, and is reserved at the 30 second maximum, then corrected down to the delivered length after generation. Gemini Omni Flash ignores this when a video reference is selected because the selected video supplies the timing.
`aspect_ratio`	`"auto" \| "21:9" \| "16:9" \| "4:3" \| "1:1" \| "3:4" \| "9:16"` for Seedance 2; `"16:9" \| "9:16"` for Gemini Omni Flash; `"auto" \| "16:9" \| "4:3" \| "1:1" \| "3:4" \| "9:16"` for Wan 3.0	No	`"auto"` for Seedance 2; `"16:9"` for Gemini Omni Flash; `"auto"` for Wan 3.0	Output aspect ratio. Seedance 2.5 accepts the same values as Seedance 2, except image-to-video, which always uses the source image aspect ratio. Wan 3.0 maps `"auto"` to the provider’s adaptive aspect ratio. Gemini Omni Flash ignores this when a video reference is selected because the selected video supplies the aspect ratio.
`generate_audio`	`boolean`	No	`true`	Seedance 2 and Seedance 2.5 only. Generate synchronized audio when supported. MiniMax H3 and Wan 3.0 always generate native audio.
`selected_context`	`{images:string[],videos:string[],audio:string[]}`	No	empty lists	Selected upstream references. Seedance 2 supports image, video, and audio references. Gemini Omni Flash supports up to 10 selected images or one selected video; audio is not supported. MiniMax H3 supports up to 9 images, 3 videos, and 3 audio clips (12 total), and audio references require at least one image or video. Wan 3.0 supports up to 10 images, 5 videos, and 5 audio clips (20 total), with video and audio references limited to 15 seconds each. Seedance 2.5 supports up to 30 images, 10 videos, and 10 audio clips (50 total), and audio references require at least one image or video.

Example

{
  "prompt": "A cinematic macro shot of coffee pouring into a glass mug",
  "model": "seedance-2",
  "resolution": "1080p",
  "duration": "8",
  "aspect_ratio": "16:9",
  "generate_audio": true
}

For 4K output, use:

{
  "prompt": "A cinematic macro shot of coffee pouring into a glass mug",
  "model": "seedance-2",
  "resolution": "4k",
  "duration": "8",
  "aspect_ratio": "16:9",
  "generate_audio": true
}

4K Seedance generations are billed at 2x normal Seedance.For Gemini Omni Flash with image references, use selected images and set duration/aspect ratio:

{
  "prompt": "Animate the selected product image into a slow cinematic turntable shot",
  "model": "gemini-omni-flash",
  "resolution": "1080p",
  "duration": "8",
  "aspect_ratio": "16:9",
  "selected_context": {
    "images": ["upstream-image-context-key"]
  }
}

For MiniMax H3 text-to-video, set the resolution, the duration in seconds, and an aspect ratio. Output always includes native audio.

{
  "prompt": "A cinematic drone shot over a foggy pine forest at dawn",
  "model": "minimax-h3",
  "resolution": "2K",
  "duration": "6",
  "aspect_ratio": "16:9"
}

For MiniMax H3 reference-to-video, select subject images, video clips, and audio clips to keep consistent across the generation. Any selected context routes to this mode.

{
  "prompt": "@Image1 walks through the set from @Video1 while @Audio1 plays",
  "model": "minimax-h3",
  "duration": "8",
  "aspect_ratio": "auto",
  "selected_context": {
    "images": ["subject-image-context-key"],
    "videos": ["motion-video-context-key"],
    "audio": ["audio-context-key"]
  }
}

MiniMax H3 is billed at 5, 6, 13, or 16 credits per output second at 480p, 768p, 2K, or 4K, plus the same per-second rate for each reference-video second and 4 credits for each reference image beyond the first five.For Seedance 2.5, set the resolution and a duration up to 30 seconds. Selected references route the request automatically: no references runs text-to-video, exactly one image runs image-to-video (which always uses the source aspect ratio), and anything else runs reference-to-video. Each reference video or audio clip must be at least 1.8 seconds; selected video references are trimmed to 30 seconds combined, and audio references are trimmed the same way.

{
  "prompt": "@Image1 walks through a neon-lit market in one continuous 30 second take",
  "model": "seedance-2.5",
  "resolution": "720p",
  "duration": "30",
  "aspect_ratio": "16:9",
  "generate_audio": true,
  "selected_context": {
    "images": ["subject-image-context-key"],
    "videos": ["motion-video-context-key"]
  }
}

Seedance 2.5 is billed at 23.11 credits per 720p second and 10.28 credits per 480p second, discounted to 60% of those rates when a video reference is selected. With video references, reference-video seconds are billed alongside output seconds; image and audio references are not billed. An "auto" duration is charged for the 30-second maximum up front and adjusted down to the delivered video’s length once the generation finishes.For Gemini Omni Flash with video reference, select one upstream video. The selected video input uses the first 10 seconds; duration and aspect ratio come from that video, and resolution still applies to the edited output.

{
  "prompt": "Change the background to a bright studio while keeping the subject motion",
  "model": "gemini-omni-flash",
  "resolution": "720p",
  "selected_context": {
    "videos": ["upstream-video-context-key"]
  }
}

Getting Started

2. Quickstart

3. Mosaic Philosophy

4. Tiles

5. Canvas

6. Models

7. How To

8. More

How It Works

API Usage

API Info

Node params

Example

​How It Works

​API Usage

​API Info

​Node params

​Example

How It Works

API Usage

API Info

Node params

Example