Veo 3.1 last-frame control: which Sume video models take an end frame

Sume does not list Veo 3.1 in its video catalog as of 2026-10-08, so you cannot call it through Sume. If you need first-and-last-frame video there, the models that accept an end frame are Seedance 2.5, 2.0, Fast and Mini, Kling Video v3 Pro, Wan 3.0, MiniMax H3, MiniMax H3 Max and Gemini Omni Flash 1.1. Grok Imagine Video 1.5, Genjutsu and H3 Max Recast do not.

Google's Gemini API video page presents Veo 3.1 as a model for native-audio video that supports video extension, frame-specific generation and image-based direction, and says to use it when scene extension or last-frame control is required (Google page, read 2026-10-08). Sume's catalog does not include a Veo scene-extension equivalent; it offers end-frame control only.

How to send both frames

On POST /v1/videos, send frame_images with one entry marked first_frame and one marked last_frame. Each entry needs a frame_type. On the Video 1.0 and Video Router shapes use image_url and end_image_url; the end frame needs the first frame to be present too.

Each model lists the frame_type values it takes in supported_frame_images; read that field before you submit.

{
  "model": "wan-3.0",
  "prompt": "The door opens and the room fills with light",
  "frame_images": [
    {"type": "image_url", "image_url": {"url": "https://example.com/closed.png"}, "frame_type": "first_frame"},
    {"type": "image_url", "image_url": {"url": "https://example.com/open.png"}, "frame_type": "last_frame"}
  ],
  "resolution": "720p",
  "duration": 5
}

What it costs

The frames add no line item. A 5-second clip costs $0.625 on Wan 3.0 at 720p, $0.50 on MiniMax H3 Max at 768p, $1.05 on Kling with audio and $0.625 on Gemini Omni Flash 1.1 at 720p, which all use per-second prices. Seedance 2.5 at 720p is $2.889 for the same length.

Which one fits a first-to-last transition

Transitions need both ends to match, so prefer a model with a wide aspect and resolution range so you can match your assets without cropping: Seedance covers 21:9 to 9:16 and three resolutions, Wan 3.0 covers 2 to 30 seconds. Kling Video v3 Pro takes 16:9, 9:16 and 1:1 only, so square or vertical assets need to be exact. If your end frame is a different aspect than your first frame, resize before upload.

Chaining clips with end frames

Because every model in the table takes an end frame, you can chain: use the last frame of clip A as the first frame of clip B and name a planned last frame for B. This gives continuity across segments when you need more than the 30-second cap. It is a manual extension, not a native one: each clip is a separate job with its own price.

• Pick the same resolution and ratio for every segment.
• Send an Idempotency-Key per segment.
• Use callback_url to start the next segment when one completes.