Krea 2 style references and moodboards vs Sume reference images

Krea 2 style transfer extracts the style of one or many reference images, and a moodboard can hold dozens. On Sume, references are image URLs on input_references, public HTTPS only, and each model caps how many it takes.

Krea facts are from its Krea 2 API overview; Sume facts are from Image models, read 2026-10-01.

What do style references and moodboards mean on Krea 2?

How do references work on Sume?

Add input_references entries of type image_url, each a public HTTPS URL. Localhost, private-network and non-HTTPS URLs are rejected before submission. The docs describe the field as reference images for image-to-image and list no separate style or moodboard role.

curl https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-image-2","prompt":"a lighthouse in the same style as the reference","input_references":[{"type":"image_url","image_url":{"url":"https://example.com/style-1.jpg"}}]}' 

How many references can I send?

It depends on the model. In the catalog code, edit-capable entries default to max_images of 4 and Grok's is 1; the docs say ChatGPT Image 2.5 takes up to 16, and some models take none. Nothing in the docs approaches a dozens-of-images moodboard, so pick your best few images.

What should I do for a style that needs many images?

Choose the reference images that carry the style most clearly, state the style in words in prompt, and test with the model's maximum. See image generation with reference images for more on the call shape.