Luma uni-1 web_search grounding vs Sume's unsupported_parameter

Luma's uni-1 and uni-1-max accept web_search: true, which searches the web for visual references before generating. Sume has no equivalent field; sending one returns 400 unsupported_parameter, and the supported way to steer a Sume image is input_references.

Luma facts are from its Models page; Sume facts are from Image models, read 2026-10-01.

What does web_search do on Luma?

The Models page lists web search grounding as a capability of both image models: when web_search is true, the model searches the web for visual references before generating. The page gives no further detail, such as which sources it uses.

What happens if I send it to Sume?

The docs say a request that sets a parameter the selected model does not list is rejected with 400 unsupported_parameter rather than silently dropped. The Image API request table has no web-search row, so the request fails instead of generating without grounding.

curl https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"sume/auto","prompt":"a 1960s Vespa","web_search":true}'
# expect: 400 unsupported_parameter

How do I supply visual references on Sume?

Find the reference images yourself and pass them as public HTTPS URLs in input_references, which the docs describe as reference images for image-to-image. Check the target model's input_references descriptor first: a model with {"min": 0, "max": 0} is text-to-image only and rejects references.

You can also put the facts you found into prompt. That keeps the research step under your control, which matters when the subject must be accurate.

Is there a grounding option on any Sume model?

Not documented in the Image API page. The post on Nano Banana search grounding covers the same question for a Google model.