Edit a high-resolution photo with AI: edit first, then upscale

Sume does not promise to return an edit at your photo's full megapixel count. Image edits are requested at catalog tiers (512, 1K, 2K, 4K), so the practical route is to edit at a tier, keep the original shape with aspect_ratio: "auto", and then enlarge the result with Image Upscale 1.0.

Facts are from the Image API docs and the Sume OpenAPI document behind the API reference, read 2026-09-30. The docs do not state a maximum input size for edits, so test with one of your own files.

What output sizes does an edit give me?

The resolution field is a tier: 512, 1K, 2K or 4K. A model only accepts the tiers its catalog descriptors list, so read supported_parameters for the model you choose. The docs do not map a tier to an exact pixel count for every model, so check the dimensions of what comes back instead of assuming.

How do I keep the photo's shape?

Set aspect_ratio: "auto". The docs say to prefer it on edit and image-to-image calls to match the reference, and that leaving the field out is not equivalent. Without it a portrait photo can come back in another shape.

How do I enlarge the edited image?

Send the edited image's public HTTPS URL to Image Upscale 1.0. The request needs image_url, takes an upscale_factor (default 2), and does not accept provider model ids. The page Image upscale API lists every field.

curl -X POST https://api.sume.com/v1/image-upscale-1.0/upscale \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: upscale-edited-photo-001" \
  -d '{
    "image_url": "https://media.sume.com/path/to/edited.png",
    "upscale_factor": 2
  }'

Why edit first and upscale last?

An edit regenerates pixels, so any detail it softens is baked in. Enlarging after the edit means you only enlarge the version you keep. This ordering is a workflow choice, not a rule from the docs; AI image 4K: generate or upscale compares the generate-at-4K route.

What if I need the original resolution exactly?

Edit the region you care about, then composite the result back onto the original in your own tooling. Sume's docs describe no step that writes an edit back to the full-size source, so that final merge is yours.