Higgsfield Soul on Sume: text to image, 1 or 4 images, 720p or 1080p

Soul (Higgsfield) is in Sume's image catalog as higgsfield/soul. It is text-to-image only, takes n of 1 or 4, offers 720p or 1080p as resolution, and returns PNG. It does not accept reference images and has no auto ratio, so it is the wrong tool for edits and the right one when you want a style sweep from a prompt alone. Sume's catalog describes the price as an estimate for the Sume account, not a published tariff, so read it from the endpoints call rather than from a blog post.

What the catalog says

The model row in the Sume source lists text-to-image only, num_images 1 or 4, resolution 720p or 1080p, and an output format chosen by the provider. Reference-guided work belongs to models that list input_references, which you can filter for with GET /v1/images/models (Image API).

A request

Ask for four to get a spread, then take the best one into an editing model. Sume sends 200 when the job finishes inside 30 seconds and 202 with a job envelope when it does not, so handle both (jobs and results).

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "higgsfield/soul",
  "prompt": "street portrait at dusk, 35mm film look, warm window light",
  "n": 4,
  "resolution": "1080p",
  "aspect_ratio": "3:4"
}'

When to pick something else

Need text in the image, a mask, or an edit of your own photo? Use a model that lists those capabilities, such as GPT Image 2.5 for masks and many references. Need a cheap fast draft with a reference? Compare the other rows with the capability matrix call.