12 colorway product clips: Nano Banana 2.1 recolor + Grok video, $1.95

To make a product video for every colorway, recolor one approved packshot with Nano Banana 2.1 (nano-banana-2.1), then animate each recolored still with grok-imagine-video-1.5. On Sume that is two calls per color: a $0.10 still plus a 5-second clip at $0.0625, or $0.1625 per colorway and $1.95 for twelve.

Both models were in the news this week. Google's Nano Banana 2.1 page lists multi-image fusion, object fidelity for up to 10 objects, and 1K as the default size (read 2026-10-12). xAI published a Grok Imagine Video 1.5 Lite page with text and image input and $0.020 per second output (read 2026-10-12). Lite is a different id from the grok-imagine-video-1.5 that Sume lists, so do not copy xAI's rate into your Sume budget.

What does one colorway cost on Sume?

Sume's image catalog bills nano-banana-2.1 at $0.10 for a 1K image, and its video catalog bills grok-imagine-video-1.5 at $0.0125 per second. That model takes a source image only (image-to-video), 4 to 15 seconds, at 480p or 720p. Confirm both numbers with GET /v1/images/models and GET /v1/videos/models before a large run, because the catalog is the live source of truth.

The clip is the cheap half. Stills cost eight times more per unit than one second of video, so a rejected recolor is the expensive mistake, not a rejected clip.

How do I recolor the packshot?

Send the approved packshot as an input_references item on POST /v1/images, and name the one thing that changes. Keep everything else in the prompt a constraint: same angle, same lighting, same background, same logo and label. Reference URLs must be public HTTPS.

Check each result against the original before it reaches a clip. Print on packaging, a logo, or a size label can drift when a model repaints a surface. If a label must be exact, overlay the real label afterward rather than trusting the recolor.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: mug-sage-still" \
  -d '{"model":"nano-banana-2.1","prompt":"Same mug, same angle, light and background. Change only the glaze to sage green. Keep the logo unchanged.","input_references":[{"type":"image_url","image_url":{"url":"https://example.com/mug-packshot.jpg"}}]}'

How do I animate each approved still?

Pass the approved recolor as the first frame in frame_images to POST /v1/videos with model: "grok-imagine-video-1.5" and duration: 5. The output frame follows the source image's ratio, so crop the packshot to the ratio you want before you recolor, not after. Use a unique Idempotency-Key per colorway so a retry never bills twice, and keep the prompt to one camera move.

Plan for the queue: Sume admits jobs against your plan's concurrency, and extra valid jobs wait as queued instead of failing. Twelve colorways on a plan with 4 processing slots simply finish in waves.

When is this the wrong approach?

For a catalog of one shape in many colors, the pipeline is cheap enough to run for every variant and let people see the product moving instead of reading a color name.

• The product changes shape between colors (a handle, a strap, a lid). Re-shoot or edit those, because a recolor will not add geometry.
• A color must match a physical swatch. A model recolor is an approximation; use it for listing motion, not for color proof.
• You need sound. This clip route is image-to-video only; add voice or music in a separate step.