Is Kandinsky 6.0 Video on Sume? No: query the catalog by need

No: Kandinsky 6.0 Video is not in Sume's video catalog, and nothing in the API source or docs of the Sume repository names it. Sume lists hosted ids only, so open-weights releases like this one are not callable through POST /v1/videos. What you can do is describe what you wanted from Kandinsky (sound, 5 seconds, a first frame) and filter the live catalog for ids that match.

The Kandinsky facts are from the project repository and the technical report, read 2026-10-11. Sume's request fields are from Video Generation. The list of ids changes, so treat the script at the end as the source of truth and this page as the map.

How this was checked

I searched the Sume checkout this post was written against (docs, API source, web app and packages) for the name Kandinsky and found no hit. The docs name the discovery route directly: GET /v1/videos/models returns every video id with its supported_durations, supported_resolutions, generate_audio and pricing_skus. That endpoint, not a blog table, decides what you can send today.

Kandinsky feature to Sume field

Each thing you would do with Kandinsky has a place on the Sume request, if a listed model supports it.

What differs in practice

Kandinsky's repository lists inference defaults for Pro of 50 steps and guidance 5.0. Those are knobs on your machine. Sume exposes none of them: the portable controls are prompt, duration, resolution, aspect ratio, first or last frame, reference media and the generate_audio flag. If your workflow depends on step count or guidance, that is a reason to run open weights yourself.

The other difference is lifecycle. A Sume video request is asynchronous. You get a job id, poll the polling URL or pass an HTTPS callback_url, then download from unsigned_urls. Send an Idempotency-Key so a retry does not create a second job.

Script: which listed ids fit a Kandinsky-shaped job

Run this with SUME_API_KEY set. It prints how many ids are listed, whether any id contains the word Kandinsky, and every id that can make audio and accepts a 5-second clip.

import os
import requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
r = requests.get("https://api.sume.com/v1/videos/models", headers=H, timeout=30)
r.raise_for_status()
ids = [m["id"] for m in r.json()["data"]]
print(len(ids), "video ids")
print("kandinsky listed:", any("kandinsky" in i.lower() for i in ids))
for m in r.json()["data"]:
    if m.get("generate_audio") and 5 in (m.get("supported_durations") or []):
        print(m["id"], m["supported_resolutions"])

If the list is empty or changes

If no id passes the filter, relax one constraint at a time. Drop the audio filter and add sound later with the audio tools, or allow 6 to 10 seconds and trim. Do not assume a model has a feature because a neighbor does; the docs call out, for example, that Gemini Omni Flash 1.1 takes 3 to 10 seconds, not the 15 seconds that other catalog ids top out at.