Leonardo names its reference presets instead of one similarity dial

The image

The documents are two of Leonardo.Ai's own developer recipes: "Generate with Image Guidance using multiple ControlNets" and "Generate with Image to Image Guidance using Uploaded Images," both hosted at docs.leonardo.ai. Each shows a code walkthrough rather than a finished image: a presigned-upload step, a request body naming specific fields, and a resulting generation described in prose, current as of this reading on 16 September 2026.

What the documents show

The multi-ControlNet recipe names two specific preprocessor identifiers in its example request — "133" for "Character Reference Id" and "67" for "Style Reference Id" — combined in a single "controlnets" array, each with a documented "strengthType" set to "Mid" or "High," and an "initImageType" of either "UPLOADED" or "GENERATED." The separate Image to Image recipe documents a single-image path instead, governed by an "init_strength" field the page states "must float between 0.1 and 0.9," with an example value of 0.5, and explicitly restricts the request to one uploaded image per generation. Both pages note that generated results are not immediately available and recommend a webhook callback over a fixed wait, a shared operational detail rather than a claim about image fidelity.

Production context

Read editorially, offering a numbered strength range (0.1 to 0.9) alongside a named preprocessor identifier gives a production pipeline two separable levers — how much an uploaded reference constrains the output, and which aspect of that reference (character or style, per the documented IDs) is being read — closer to a compositing tool's named channels than to a single opaque "similarity" slider.

Reference versus imitation

Both recipes document conditioning on a specific uploaded or previously generated image identified by its own ID, not on an artist's name or an unnamed "style." Because the character- and style-reference identifiers are Leonardo's own numbered presets rather than descriptions borrowed from another platform, this entry does not claim they function identically to any competing tool's reference feature — only Leonardo's own documentation is cited here for how its own parameters behave.

• Was the request built with a single init_strength value, or a multi-ControlNet array with named preprocessor IDs?
• Which strengthType or init_strength value was used, and is it recorded with the output?
• Does the cited documentation describe Leonardo's own feature, or is a claim being extended to another platform without its own source?

A numbered parameter and a documented range are the kind of detail worth keeping attached to any generated reference image, since they are what the tool's own developer page — not the image itself — can actually prove about how it was made.