Check video model limits before you submit: GET /v1/videos/models

Call GET /v1/videos/models and check your duration, resolution, aspect_ratio, and reference types against the model's entry before you submit. Limits are not shared across models: on Sume seedance-2.5 accepts 4-30 seconds, wan-3.0 accepts 2-30 seconds, and most other catalog models stop at 15 seconds or less, with gemini-omni-flash-1.1 at 3-10.

The endpoint returns a data array, each entry with supported_durations, supported_resolutions, supported_aspect_ratios, supported_input_references, generate_audio, and pricing_skus.

Why validate first

Limits differ per model, so a request that is valid for one model can be invalid for another. Reading the catalog also keeps you from hard-coding a model id that is no longer listed.

The vendors announce longer clips (Seedance 2.5, Wan3.0) before an API serves them everywhere. The catalog shows what your workspace can call today.

A preflight function

The check below raises a clear error instead of a failed paid call. It does not replace server validation, which stays the final word.

import os, requests
H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
catalog = requests.get("https://api.sume.com/v1/videos/models",
                       headers=H, timeout=30).json()["data"]
by_id = {m["id"]: m for m in catalog}

def preflight(model: str, duration: int, resolution: str) -> None:
    m = by_id.get(model)
    if m is None:
        raise ValueError(f"{model} is not in the catalog: {sorted(by_id)}")
    if duration not in m["supported_durations"]:
        raise ValueError(f"{model} durations: {m['supported_durations']}")
    if resolution not in m["supported_resolutions"]:
        raise ValueError(f"{model} resolutions: {m['supported_resolutions']}")

preflight("seedance-2.5", 30, "720p")
print("ok")

Cache it, but not forever

Fetch the list at startup and refresh it on an interval, say hourly. A new model appears in the response without a code change on your side.