Brand character keyframe: GPT Image 2.5 refs, then first_frame

For a brand character that looks the same in every video, make the opening frame as an image first. GPT Image 2.5 on Sume accepts up to 16 reference images, so a model sheet and a product shot can steer one keyframe. Then send that keyframe to POST /v1/videos as a first_frame in frame_images, so the clip begins on the approved image.

Why start with an image

A video model that starts from text invents the character each time. A video model that starts from an approved still keeps it. An image is also cheaper to redo than a clip, so you can reject five keyframes before you spend on one video.

Sume lists openai/gpt-image-2.5 (Flare) and openai/gpt-image-2.5-sunburst. Both support text-to-image, up to 16 image references, an optional mask_url and background: auto|transparent|opaque. The quality field takes auto, low, medium, high, xhigh or max, and the default is high.

Custom sizes and limits

image_size takes named presets, auto or custom pixels. For custom pixels, both edges must be multiples of 16 and the maximum edge is 3840. The ratio must be at most 3:1, and the image must hold between 655,360 and 8,294,400 pixels. A 1080x1920 vertical frame fits: both edges divide by 16 and the pixel count is 2,073,600.

Animate the approved frame

After you approve the keyframe and have a public HTTPS URL for it, pin it as the first frame. The catalog tells you which models accept first_frame, so check supported_frame_images for the model you choose.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: mascot-spot-001" \
  -d '{
    "model": "seedance-2",
    "prompt": "The character waves, then holds up the product and smiles",
    "duration": 6,
    "aspect_ratio": "9:16",
    "frame_images": [
      {"type": "image_url",
       "image_url": {"url": "https://example.com/mascot-keyframe.png"},
       "frame_type": "first_frame"}
    ]
  }'

Keep the character stable

Use the same model sheet in every keyframe request, and keep the character description out of the video prompt. The picture already carries it. Spend the prompt on motion. When a new campaign needs a new outfit, change the references, not the whole workflow.