House cleaning before-and-after reel with the sume-before-after Format

To make a before-and-after reel for a cleaning job, call the catalog Format sume-before-after with the before photo, the after photo, and a short instruction. Sume describes it as producing a finished before-and-after video with matched identity, framing and lighting and a clear reveal, and its description names cleaning as a use.

The Format starts from your evidence. It is built for a transformation that your photos support, so the honest limit is simple: if the after photo does not show a clean room, no tool should be asked to make one up.

Why this format fits a cleaning business

Cleaning, detailing and repair trades sell a visible result. Short clips with synchronized audio and a clear subject are the practical range for polished AI video right now, and a roundup of this month's trends puts the short-form sweet spot at 5 to 30 seconds (AI Video Generation Trends, read 2026-10-07).

A before-and-after is the plainest version of that: two states of one room, one reveal, a caption with your service area. The Format handles the composition; you supply the proof.

What to send

A Format run is one POST to /v1/formats/sume/sume-before-after/runs. Put your data in input, attach up to 30 images, and write the task in instruction (up to 8000 characters). Every create needs an Idempotency-Key, and a key with the formats:write scope.

Shoot both photos from the same spot, at the same height, in similar light. The Format is built to match framing and lighting, but a before taken from the doorway and an after taken from the window gives it nothing to match.

curl -sS -X POST "https://api.sume.com/v1/formats/sume/sume-before-after/runs" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: kitchen-job-1042" \
  -d '{
    "instruction": "Kitchen deep clean. Vertical 9:16. Show the counter and sink before, then the reveal. No price on screen.",
    "input": { "business_name": "Example Cleaning Co", "service": "kitchen deep clean" },
    "attachments": [
      { "type": "input_image", "image_url": "https://example.com/kitchen-before.jpg" },
      { "type": "input_image", "image_url": "https://example.com/kitchen-after.jpg" }
    ],
    "generation_spend_cap_usd": 10
  }'

Cap the spend and read the receipt

generation_spend_cap_usd is the ceiling for one run and the platform maximum is $500, so a small business can set a number it is comfortable losing on a bad take. The 202 response gives status_url and result_url; poll status, then fetch the result for the video. A repeat of the same Idempotency-Key with the same body returns the original run, not a second charge.

To change the take, continue the same run with previous_run_id and a new instruction (for example, shorter, or without the logo). That keeps the thread together instead of starting from scratch.

What it cannot do

A Format does not verify that the after photo is real, and Sume does not post the reel for you. If your market has rules about before-and-after claims, use real photos from the real job and keep the originals. Read the Format's page with GET /v1/formats/sume/sume-before-after before you rely on it, because the io profile tells you what input shape it expects.