Virtual try-on or virtual fitting: which Sume Format for apparel?

Use sume-virtual-try-on for an ad, where a creator changes into the garment and the reveal is the hook. Use sume-virtual-fitting for a product page or a returns-aware preview, where the point is to show silhouette, sleeve length and drape with a small turn. Both take a garment still as the clothing reference, both make a 9:16 video, and the shipped examples build the first frame with ChatGPT Image 2 and animate it with Seedance 2.5 reference-to-video.

How do they differ?

From the Format catalog and shipped example briefs, read 2026-10-01:

How do I call either one?

The wire contract is the same, only the slug differs. Send the garment as an attachment and name the garment's real colour, cut and length in the brief, since that is what the fitting Format is meant to show.

curl -X POST https://api.sume.com/v1/formats/sume/sume-virtual-fitting/runs \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: fitting-shirt-001" \
  -d '{
    "instruction": "9:16 fitting preview of this oversized shirt, show sleeve length and drape, small turn",
    "attachments": [{ "type": "input_image", "image_url": "https://example.com/shirt.jpg" }],
    "generation_spend_cap_usd": 20
  }'

What are the limits?

• A fitting preview is generated, so it is not a size guide. It cannot tell a shopper how the garment fits their body, and it should not be presented that way.
• Fabric drape and sleeve length are the first things to check against the real garment.
• If the garment has a printed pattern or text, check it frame by frame.
• For a catalogue of garments, queue them as a bulk run, but the queue has no webhook; poll it, and read counts.failed when it reports completed.
• Using a real person's likeness needs their permission.

How do I tell them apart before I spend?

Call GET /v1/formats/sume/sume-virtual-try-on and GET /v1/formats/sume/sume-virtual-fitting. Each returns a description and an io profile with input_kind and output_kind, plus a showcase of an output the Format actually produced. Look at the two showcases side by side; they show the difference faster than a description.

Which one first for Q4?

Run the ad version on your best-selling gift item and the fitting version on the item with the most returns questions. Compare them by what each is for, not by one score. When the clips are approved, the same slugs can be queued for the rest of the line.