Skip to main content
Glama

Image To Video

image_to_video

Generate video from static image(s), or from reference images, videos and audio.

Two modes, mutually exclusive:

  • Keyframes: first_frame_image (and optionally last_frame_image) is animated.

  • References: ref_images / ref_videos / ref_audios, on models that support them (the model list shows "refs: ..." for those). The prompt addresses each reference by its tag: <Picture 1>, <Picture 2> for images, <Video 1> for a video, <Audio 1> for audio, numbered from 1 within each type in the order given. Order matters: swapping two images swaps what the tags refer to. Example prompt: "<Picture 1> walks into the room from <Picture 2>, speaking with the voice of <Audio 1>".

Accepts files as:

  • Public URLs (recommended: chains with text_to_image, and the ONLY way in for a file of your own - chat clients do not pass files through to tools)

  • Data URIs or raw base64 (for callers that already hold the bytes)

IMPORTANT: Check model specifications using get_available_models() before calling. Pay attention to limits for width/height/frames/fps and, in reference mode, the max_ref_* counts and durations.


Available models:

  • Ltx2_5_22B_Dist_INT8: size=512-1344x512-1344, fps=24 (fixed), frames=49-241, w/h rounded down to a multiple of 64, ≤1.03Mpx, side≤1344, aspect≤2, audio2video audio=1-10s, no negative_prompt

  • MiniMaxH3_33B_Turbo_INT8: size=512-1344x512-1344, fps=24 (fixed), frames=56-243, w/h rounded down to a multiple of 32, ≤1.03Mpx, side≤1344, refs: ≤3 images, ≤1 video (0.5-2.3s), ≤3 audio (1-10s), ≤3 files total, no negative_prompt

  • Ltx2_3_22B_Dist_INT8: size=512-1024x512-1024, fps=24 (fixed), frames=49-241, audio2video audio=1-11s, no negative_prompt

  • Ltxv_13B_0_9_8_Distilled_FP8: steps=1 (fixed), size=256-768x256-768, fps=30 (fixed), frames=30-120

  • kling-4 (test model): size=720-2520x720-1920, fps=24 (fixed), frames=73-721, w/h rounded down to a multiple of 2, ≤2.72Mpx, aspect≤2.34, no negative_prompt

  • seedance-2.5 (test model): size=432-2206x432-2206, fps=24 (fixed), frames=97-721, w/h rounded down to a multiple of 2, ≤2.09Mpx, side≤2206, aspect≤2.34, refs: ≤9 images, ≤1 video (1.8-30.2s), ≤3 audio (1.8-30.2s), ≤12 files total, no negative_prompt Models marked (test model) are experimental: they may change or disappear. Prefer an unmarked model unless the caller asks for one of them.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "prompt": "<prompt>",
      "model": "<model>"
    },
    "name": "image_to_video"
  }
}

Input Schema

NameRequiredDescriptionDefault
fpsNoFrames per second. Defaults to the model's own value (see the model list); most models fix it
seedNoRandom seed for reproducibility
modelYesVideo generation model name
stepsNoNumber of inference steps. Defaults to the model's own value (see the model list); ignored by models that do not support steps
widthNoVideo width in pixels. Defaults to the model's own value (see the model list)
framesNoNumber of video frames. Defaults to the model's own value (see the model list); each model has its own range
heightNoVideo height in pixels. Defaults to the model's own value (see the model list)
promptYesText prompt for video generation
ref_audiosNoReference audio clips (URL, data URI or base64), in order: ref_audios[0] is <Audio 1> (or <Audio 2> after a sounded reference video). Needs at least one ref_images or ref_videos entry
ref_imagesNoReference images (URL, data URI or base64), in order: ref_images[0] is <Picture 1> in the prompt, ref_images[1] is <Picture 2>, and so on. Each within a 1:4-4:1 aspect ratio. Only for models listing image refs in the model list; replaces first_frame_image
ref_videosNoReference videos (URL, data URI or base64), in order: ref_videos[0] is <Video 1>. Usually at most one, a few seconds long (see the model list). A video with a soundtrack takes <Audio 1> and shifts every ref_audios label by one
webhook_urlNoHTTPS URL for deAPI to call when this job finishes. Notifies your own systems; the tool still returns the result here
enhance_promptNoLet the API rewrite the prompt before generating. Only on models with a prompt-booster guide (422 otherwise); its fee is charged on the job and is not in the /price quote
guidance_scaleNoGuidance scale. Leave unset to use the model's own default; no model in the catalogue currently supports guidance
webhook_secretNo32-255 char key deAPI signs the callback with (HMAC-SHA256). Needs webhook_url. A secret given in chat stays in the transcript
negative_promptNoThings to exclude (optional)
last_frame_imageNoLast frame image (optional, URL, data URI, or base64). Not allowed with reference inputs
first_frame_imageNoFirst frame image as URL (e.g., from text_to_image result), data URI (data:image/png;base64,...), or base64 string. URLs are recommended when chaining from text_to_image to avoid base64 context bloat. Required unless ref_images or ref_videos are given; the two modes are mutually exclusive

Output Schema

NameRequiredDescriptionDefault

No arguments

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed3 schema fields changed
    • changedInput schema / properties / ref_audios / anyOf
      Previous value: -[
      -  {
      -    "items": {
      -      "type": "string"
      -    },
      -    "type": "array"
      -  },
      -  {
      -    "type": "null"
      -  }
      -]New value: +[
      +  {
      +    "items": {
      +      "type": "string"
      +    },
      +    "maxItems": 16,
      +    "type": "array"
      +  },
      +  {
      +    "type": "null"
      +  }
      +]
    • changedInput schema / properties / ref_images / anyOf
      Previous value: -[
      -  {
      -    "items": {
      -      "type": "string"
      -    },
      -    "type": "array"
      -  },
      -  {
      -    "type": "null"
      -  }
      -]New value: +[
      +  {
      +    "items": {
      +      "type": "string"
      +    },
      +    "maxItems": 16,
      +    "type": "array"
      +  },
      +  {
      +    "type": "null"
      +  }
      +]
    • changedInput schema / properties / ref_videos / anyOf
      Previous value: -[
      -  {
      -    "items": {
      -      "type": "string"
      -    },
      -    "type": "array"
      -  },
      -  {
      -    "type": "null"
      -  }
      -]New value: +[
      +  {
      +    "items": {
      +      "type": "string"
      +    },
      +    "maxItems": 16,
      +    "type": "array"
      +  },
      +  {
      +    "type": "null"
      +  }
      +]
  2. First observed

TDQS

A4.6/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations supplied, the description carries the full behavioral burden and does so thoroughly: it explains mode exclusivity, reference tag ordering, file input forms, model-dependent limits, reference support, and experimental-model caveats. These are material behaviors an agent needs before invoking the tool.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is front-loaded with purpose and modes, then file handling, then model limits. It is long but largely justified for an 18-parameter model-dependent tool; the model catalogue is detailed and could be partially deferred to get_available_models, which keeps it from a 5.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a complex generation tool with 18 parameters, model-dependent constraints, no annotations, and an output schema, the description covers the essential invocation context: modes, input sources, reference semantics, model limitations, and prerequisite model checking. Return values are covered by the output schema, so no major completeness gap remains.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so the schema already documents most parameters well. The description adds meaningful semantic guidance beyond the schema, especially model-specific width/height/frames/fps/reference limits, reference tag mapping and ordering, mutual exclusivity, and URL chaining advice, though some of this repeats schema-level details.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource: generate video from static image(s) or from reference images, videos, and audio. It also defines two mutually exclusive modes, so an agent can identify this as the image-conditioned video generator rather than a generic text-to-video tool.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives clear internal usage guidance: two mutually exclusive modes, reference-tag addressing, public URL recommendation for chaining, and an explicit instruction to check get_available_models before calling. It does not explicitly compare against siblings such as text_to_video or audio_to_video or state when not to use this tool, so it falls short of a 5.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.