Midjourney image weight returns in V8.1: reference influence on Sume

Midjourney's V8.1 Alpha post lists "image prompts with weighting" among the community favourites it restored. The Sume image API has no numeric weight on a reference. If you ask how to make a reference count for more or less, the answer is to say it in the prompt, to choose fewer and cleaner references, and to pick a model whose reference limit suits the job.

What Midjourney says

From the V8.1 Alpha post.

What the Sume request offers

In the Image API docs, references go in input_references as an array of image_url entries, which must be public HTTPS. The array has no weight property. A model whose input_references descriptor is {"min": 0, "max": 0} is text-to-image only and rejects references.

What you control is the count, the order, and the prompt words that describe each reference's role.

Four ways to dial influence

Use these in combination:

• Fewer references: one reference pulls harder than five. Add references only when you need each one's contribution.
• Name the role: write "use image 1 for the pose and image 2 only for the colour palette". Numbering references in the prompt is covered in a related post.
• Describe the carry-over: "keep the composition, change the materials" versus "keep the colours, change everything else".
• Pick the model: reference limits differ by model, so read supported_parameters.input_references before you plan a many-reference prompt.

A prompt that weights in words

A phrase such as "Match image 1's layout closely. Take only the lighting mood from image 2, lightly." does the job a weight number would. It is less precise than a number, so test two or three phrasings and compare. Because there is no seed, repeat each phrasing once or twice before you decide.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Match image 1 layout closely. Take only the lighting mood from image 2, lightly.",
    "aspect_ratio": "auto",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/layout.jpg"}},
      {"type": "image_url", "image_url": {"url": "https://example.com/mood.jpg"}}
    ]
  }'

Honest limit

Words are a soft control. If you need a hard, repeatable blend ratio, the Sume image API cannot give you one today. Plan on a review pass and keep the good result's prompt next to its output.