Kling 4.0 Flash is first-frame only; what Sume's Kling 3 accepts

Kling 4.0 Flash, in early access since September 28, is a narrower model than the Kling 4.0 headline suggests. fal's explainer describes it as 20 seconds at most, 720p, 8-bit SDR and first-frame image only, which is not full image-to-video (read 2026-10-04).

Sume does not list Kling 4.0. It ships kling-3: 4 to 15 seconds, 720p or 1080p, at $0.14 per second billed with audio off and $0.21 with audio on.

What does Kling 4.0 promise, and what is available now?

Kling's own announcement says the full launch comes in October, with 3 to 30 second clips, 4K output and a larger reference system. It lists 10-bit HDR as coming soon (read 2026-10-04). The Flash tier is the part people can use first, and it carries the limits above.

So the choice today is between an early-access tier with a first-frame input and a model that is routable now.

What does kling-3 take on Sume?

Kling 3 on Sume is text-to-video and image-to-video, with frame_images for a first frame and an optional last frame. It takes no reference images, videos or audio. Aspect ratios are 16:9, 9:16 and 1:1. The rate is not tiered by resolution, so 720p and 1080p cost the same per second.

A 5-second clip is $0.70 silent and $1.05 with audio.

Which Sume models take references if you need them?

If the shot needs a subject held steady, use a row that accepts input_references: Wan 3.0, the Seedance ids or the MiniMax rows. Read supported_input_references for the exact types each one takes before pinning.

• Need a clip longer than 15 seconds: wan-3.0 or seedance-2.5 (30 seconds).
• Need a last frame: kling-3 with frame_images.
• Need image-to-video only, no audio, at 480p or 720p: grok-imagine-video-1.5.

How do you send a first-frame request?

This pins kling-3 with a first frame. The image URL is read from the environment.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/videos",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={
        "model": "kling-3",
        "prompt": "The camera pushes in as the door opens",
        "duration": 5,
        "resolution": "1080p",
        "frame_images": [{
            "type": "image_url",
            "image_url": {"url": os.environ["FIRST_FRAME_URL"]},
            "frame_type": "first_frame",
        }],
    },
)
print(r.status_code, r.json())