Product turntable ad: which Sume video models take two frames

For a product turntable ad, send a front photo as the first_frame and a back or side photo as the last_frame, and let the model fill the turn between them. On Sume's video catalog, gemini-omni-flash-1.1, wan-3.0, kling-3, minimax-h3, minimax-h3-max and the Seedance family accept an end frame; grok-imagine-video-1.5 does not, because it is an image-to-video row that takes a first frame only.

The catalog is the source here: Video Generation and the Video Router guide, both read on 2026-10-05. Pick by clip length, resolution and price, because every model that takes two frames does the same job in different limits.

Models that take a first and a last frame

The price column is the 720p or nearest native rate per second on Sume. Sume bills the provider list price times 1.25 and rounds the job up to the cent, so a one-second figure here is a rate, not a charge.

What a 5-second turn costs

A 5-second turn at 720p is $0.625 on gemini-omni-flash-1.1 or wan-3.0, $0.70 on kling-3 with audio off, and $0.375 on minimax-h3 at its native 768p. A turntable needs no sound, so generate_audio: false on Kling is the cheaper setting; Omni and H3 always make audio and you pay for it either way.

Draft first: Wan at 480p is $0.0625 a second, so a 5-second draft of the turn is about $0.32. Approve the angle pair, then pay for 1080p once.

Request

Two frame_images entries, one per frame_type. Both must be public HTTPS URLs.

import os, time, requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
payload = {
    "model": "wan-3.0",
    "prompt": "Slow 180 degree turntable of a ceramic mug on a white sweep, soft studio light",
    "duration": 5,
    "resolution": "480p",
    "aspect_ratio": "1:1",
    "frame_images": [
        {"type": "image_url", "image_url": {"url": "https://example.com/mug-front.png"}, "frame_type": "first_frame"},
        {"type": "image_url", "image_url": {"url": "https://example.com/mug-back.png"}, "frame_type": "last_frame"},
    ],
}
job = requests.post("https://api.sume.com/v1/videos", headers=H, json=payload).json()
while True:
    time.sleep(30)
    s = requests.get(job["polling_url"], headers=H).json()
    if s["status"] in ("completed", "failed", "cancelled"):
        break
print(s["status"], s.get("unsigned_urls"), s.get("usage"))

Limits to plan around

Square output rules out Omni, which takes only 16:9 or 9:16. Kling takes 16:9, 9:16 and 1:1. No catalog model accepts seed, so a rerun gives a new take, not the same turn.

If the product has text on the label, check the last frame of the result first; the end frame is a target the model approximates, not an exact pixel match.