Native 4K AI video API: LTX-2.5 claims 4K, what Sume offers instead

LTX-2.5 (Lightricks, 2026-08-11) is listed by Magic Hour's tracker, read 2026-10-06, as going up to 4K with an audio-to-video mode. It is not in Sume's video catalog, so you cannot call it here. Sume has two ways to get a 4K file: gemini-omni-flash-1.1, which renders 3-10 second clips at 4K, and minimax-h3, whose 2K and 4K are upscales of a native 768p render that Sume bills if you request them.

The vendor claim is on Magic Hour's tracker. The Sume side comes from the video generation docs and Video Router docs. Nothing here asserts that LTX-2.5's 4K is native or upscaled; the tracker only says up to 4K.

The 4K options on Sume

Billed price is the provider list times 1.25. Totals round up to the cent.

Which one for which job

If the 4K frame size matters more than clip length, Omni is the straight answer: it is rendered at that size, and the clip tops out at 10 seconds. If you want 15 seconds with stereo sound at a large frame, H3's 4K upscale reaches that, and the native render underneath is 768p. If you need 30 seconds, Seedance 2.5 and Wan 3.0 stop at 1080p.

Aspect ratio is a second filter. Omni takes only 16:9 and 9:16. H3 takes adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. A square 4K clip is therefore an H3 job, not an Omni one.

Request a 4K clip

The 4K value goes in resolution; Omni accepts lowercase 4k as an alias.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/videos",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={
        "model": "gemini-omni-flash-1.1",
        "prompt": "Slow dolly past a glass perfume bottle, soft window light",
        "resolution": "4K",
        "aspect_ratio": "16:9",
        "duration": 5,
    },
    timeout=60,
)
r.raise_for_status()
print(r.json()["id"], r.json()["polling_url"])

If you need LTX specifically

Run it where it is offered. For the Sume-side equivalents of its audio-driven mode, see LTX-2.5 and audio references. The full 2K and 4K cost grid is in 2K or 4K AI video on Sume.

Does 4K matter for your delivery?

Most short-form destinations never show a 4K frame. A vertical clip for a social feed is 1080 by 1920, and a 4K file is downscaled on upload. The cases where 4K pays are large-screen playback, a crop-in on a wide shot, and a master you will grade or cut later. Before you buy 4K for every shot, check the cost against the alternative: the same Omni clip at 1080p is $1.88 for 10 seconds against $3.75 at 4K, so 4K doubles the price.

Because Sume's catalog caps most models at 1080p, a pragmatic workflow is to render at 1080p, pick the shots that survive review, and re-render only those at 4K on Omni. The catalog does not accept a seed, so a re-render is a new take; keep the 1080p only if the 4K take differs in ways you do not want. LTX-2.5 may still be the right tool if native 4K from an open model is a hard requirement, and you would run it where it is offered.

Check the frame before you pay twice

Whichever 4K route you take, inspect the first and last frames of a draft at 720p or 1080p before you buy the large version. Faces, hands and on-screen text are the things that fail, and they fail at any resolution. A 4K clip of a broken hand is a more expensive broken hand. Render the cheap version, confirm the take, and then spend on the size.