St. Lucia Day Dec 13: bakery lussekatter pre-order clip from one photo

To promote lussekatter for St. Lucia Day, photograph one tray of saffron buns, animate it with an image-to-video clip, and burn your pickup date and order deadline on as caption cues. St. Lucia Day is December 13, which in 2026 is a Sunday, so a bakery needs its order cutoff the week before.

On Sume that is three small jobs: one video job from your photo as the first frame, an optional compose step if you want the logo card on screen, and a captions job with authored cues. None of it needs a camera crew, and each step returns a file you can inspect before you post.

What is St. Lucia Day, and why does it suit a bakery?

Wikipedia describes Saint Lucy's Day as observed on December 13, a festival of light in Scandinavia. In Sweden the traditional treat is the saffron-flavored S-shaped bun with raisins, and a girl chosen as Lucia leads a procession in a white gown and a crown of candles, serving coffee and the buns.

That gives a bakery a clean visual: golden buns, warm candlelight, steam. It also gives a hard date. A pre-order clip that says when the last order is taken converts better than a vague holiday greeting, and the date is the one fact you must get right.

How do you animate the bun photo?

Send the photo as a first frame. The Sume video guide documents frame_images with frame_type: first_frame for image-to-video, and the API infers the mode from which field you send. Ask for slow steam and a flicker of candlelight, not a new scene, so the buns look like your buns.

Pick the model from the catalog rather than from memory. GET /v1/videos/models lists supported resolutions, aspect ratios and durations per model. The docs state that wan-3.0 accepts 2 to 30 seconds, and gemini-omni-flash-1.1 accepts 3 to 10 seconds at 16:9 or 9:16 with native synced audio. For a vertical 6 to 8 second story clip, either fits.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"wan-3.0","prompt":"Soft steam rises from saffron buns, candle flicker, static camera","duration":6,"aspect_ratio":"9:16","resolution":"720p","frame_images":[{"type":"image_url","image_url":{"url":"https://example.com/lussekatter.jpg"},"frame_type":"first_frame"}]}'

The response is a 202 with an id and a polling URL. Poll until the job is completed, then download from the content URL.

How do you add the pickup date without a lookalike font?

Do not ask a video model to draw the date. Generated text drifts. Burn it on afterwards with the captions endpoint, which accepts cues with text, start and end for authored overlay text without speech-to-text. A silent clip with no cues fails with caption_no_speech, so always send cues for silent output.

Three cues are enough: the product line, the date, the deadline. For example, "Lussekatter, baked fresh" from 0 to 2 seconds, "Pickup Sunday, Dec 13" from 2 to 4, and "Order by Dec 9" from 4 to 6. Choose your own deadline from your baking schedule.

What about the logo card?

Timeline compose puts one hosted still and one hosted video on screen together and returns one MP4. The docs list a flat $0.02 per job, to be confirmed in the live catalog, and require an Idempotency-Key header. Both files must be on media.sume.com first, so import them before composing.

Which model should you pick, and how do you keep spend bounded?

Do not guess at model limits. The Sume video guide says to read GET /v1/videos/models before submitting, because limits differ per model: for example wan-3.0 accepts 2 to 30 seconds, seedance-2.5 accepts 4 to 30 seconds at 480p, 720p and 1080p, and gemini-omni-flash-1.1 accepts 3 to 10 seconds. The same catalog carries pricing_skus, so you can read the price of the length and resolution you want before you spend anything.

For a social story, a short clip at a modest resolution is enough; a higher tier costs more and a phone screen rarely shows the difference. Draft once at the lowest tier that looks right, check the cues and the dates, and only then decide whether a second render is worth it.

Keep the idempotency habit too. Where an endpoint requires an Idempotency-Key, reuse the same key when you retry the same request so a network error does not create a duplicate job.