Turn a finished ad still into a short clip with a first frame

To animate a finished ad still, send it as frame_images with frame_type: "first_frame" on POST /v1/videos, plus a prompt that describes the motion. Only models whose supported_frame_images lists first_frame accept it, so read the video catalog before you pin a model.

Image models keep getting better at text: Higgsfield lists FLUX 3 Image with more accurate text and native 4K output on 2026-10-01 (Higgsfield changelog, read 2026-10-02). Once a still passes review, a first frame keeps the clip anchored to that approved look.

What does the request look like?

The still must be a public HTTPS URL. If you send both frame_images and input_references, frame_images takes precedence and the request is treated as image-to-video.

import os
import requests

body = {
    "model": "seedance-2.5",
    "prompt": "Slow push-in, steam rising from the mug",
    "frame_images": [{
        "type": "image_url",
        "image_url": {"url": "https://example.com/approved-still.png"},
        "frame_type": "first_frame",
    }],
    "duration": 6,
    "resolution": "720p",
}
r = requests.post(
    "https://api.sume.com/v1/videos",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json=body,
    timeout=60,
)
print(r.status_code, r.json())

How do I know a model accepts a first frame?

Call GET /v1/videos/models and read supported_frame_images for the model. Do not assume: the docs show the field per model, and only models that list a type accept it. The same row lists supported_durations and supported_resolutions.

What should the first frame look like?

Match the aspect ratio you want for the clip. The video catalog lists the ratios each model accepts, such as 16:9, 9:16, and 1:1. Keep the still free of text you cannot afford to see drift, because the video model animates the whole frame. After the job completes, download from the content URL and review the clip next to the still. See Video generation for the fields.