List Video Models

GET /v1/videos/models No API key needed View .md

The video models currently served, in OpenRouter's VideoModel shape (a data list, no pagination; OpenRouter's SDKs parse it) with the durations, sizes, inputs and passthrough parameters each accepts, plus our own fields (inputs, operations, supported_fps, frame_grid). Requests are checked against these entries before they are queued. A model without a valid capability spec is not listed.

Guides: Write prompts for MiniMax-H3

Responses

application/json · video-models-payload

1 field

data array of objects
30 fields
- id string required
- canonical_slug string required
  
  The model id.
- name string required
- description string
- hugging_face_id string or null
- created integer required
- supported_durations array of integers required
- supported_resolutions array of strings required
- supported_aspect_ratios array of strings required
- supported_sizes array of strings required
- supported_frame_images array of strings or null required
- generate_audio boolean required
  
  Whether the model generates audio with the video.
- generates_audio boolean deprecated
  
  The old name of generate_audio, with the same value, kept for a deprecation period. Use generate_audio.
- seed boolean required
  
  Whether a seed is honoured.
- upscale_factor object or null required
  
  For upscaling models; always null here.
- creativity array of integers or null required
  
  For upscaling models; always null here.
- allowed_passthrough_parameters array of strings required
- pricing_skus object
  
  USD per output second as strings as the catalog writes them ("0.06"). generate_per_second and reference_per_second are the default (Turbo) price; <tier>_per_second (standard_per_second, max_per_second, ultra_per_second) each quality tier's, and <resolution>_per_second (2K_per_second, 4K_per_second) a resolution billed above the preset.
- operations array of strings
- edit object or null
  
  What POST /v1/videos/edits takes for the model; null for a model that does not edit.
  4 fields
  - resolutions array of strings required
    
    The resolutions it edits at.
  - mask array of strings required
    
    What the mask may be, an image (every frame) or a video (frame by frame).
  - audio array of strings required
    
    The soundtrack choices, keep (the source's, copied) or generate.
  - motion_guide boolean required
    
    Whether motion_guide is offered.
- inputs object
  
  For each role, how many of each media type.
- supported_fps array of integers
- frame_grid string
  
  How duration is rounded to frames, e.g. 17n+5.
- quality_tiers array of objects or null
  
  The values of quality, null for a model without tiers.
  6 fields
  - name string
  - steps integer or null
    
    The base steps the tier runs; null for the tier that runs each mode's own Turbo schedule.
  - step_range object or null
    
    The num_inference_steps counts billed at the tier, first to last; null for the Turbo tier.
    
    2 fields
    
    min integer
    
    max integer
  - multiplier number
    
    Billed seconds per output second.
  - aliases array of strings
  - price_per_second string or null
- prompt_max_length integer or null
  
  The most characters prompt may have.
- prompt_layout string or null
  
  The name of the layout the model was trained to read its prompt in, for clients that write prompts in it; null for a model without one. minimax-h3 is MiniMax's labelled fields (see Write prompts for MiniMax-H3). A prompt that does not follow it is still accepted.
- defaults object
  
  What a request that leaves them out gets.
  4 fields
  - duration integer
  - aspect_ratio string
  - resolution string
  - generate_audio boolean
- modes array of objects
  
  The model's modes in the order they are tried; the first whose when holds is the request's.
  3 fields
  - mode string
  - when object
    
    any_inputs: at least one of these media selects the mode (frames win over references). Empty for the last mode, which takes the rest.
  - schedule_steps integer or null
    
    The step count the mode's own schedule runs, which is what quality turbo and no step count mean in it; null without a tuned schedule.
- resolution_rules object
  
  The resolutions that are not a plain preset: upscale_factor (the engine enlarges the preset), upscaler and upscaler_factor (SeedVR2 enlarges the engine's clip by that factor after the render; minimax-h3's 4K, free), bill_multiplier (billed seconds per output second) and schedule_only (only the mode's own schedule runs there; another step count or tier is a 400).
- media object
  
  How each media type the model reads is taken, as an https link or a data URI.

curl https://api.vultrinference.com/v1/videos/models

import requests

response = requests.get(
    "https://api.vultrinference.com/v1/videos/models",
)
print(response.json())

const response = await fetch("https://api.vultrinference.com/v1/videos/models");
console.log(await response.json());

Response · 200

{
    "data": [
        {
            "id": "string",
            "canonical_slug": "string",
            "name": "string",
            "created": 0,
            "supported_durations": [
                null
            ],
            "supported_resolutions": [
                null
            ],
            "supported_aspect_ratios": [
                null
            ],
            "supported_sizes": [
                null
            ],
            "supported_frame_images": [
                "first_frame"
            ],
            "generate_audio": false,
            "seed": false,
            "upscale_factor": {},
            "creativity": [
                null
            ],
            "allowed_passthrough_parameters": [
                null
            ],
            "prompt_max_length": 8000,
            "prompt_layout": "minimax-h3"
        }
    ]
}