CartVisualCreate account

Leonardo Omni References or Legacy ControlNet for a Product Scene?

Use current Omni image roles for supported multi-reference composition; legacy ControlNet controls are model-dependent and slated to leave, while trained Elements are a separate legacy artifact.

Published · By CartVisual editorial team

Choose the mechanism before choosing the reference

A merchant has a verified mug photograph, a linen napkin photograph and a mood image. An old Leonardo tutorial asks for Edge or Character Reference, while the current screen offers image references and a natural-language instruction. These are not interchangeable inputs. Identify the selected model and interface first; do not troubleshoot a missing legacy slider by repeatedly uploading the same mug.

MechanismInput and controlMerchant decisionBoundary
Current Omni referencesUp to six images, described relationships and rolesCombine product, prop and scene guidanceNot a pixel-preservation lock
Legacy Image Guidance / ControlNetSpecific Edge, Pose, Depth, Style or other supported guidance; legacy premium workflow describes up to four references and independent weightsUse a particular structural control only when the model and account still expose itLeonardo warns Legacy mode will leave the platform
Trained Element / LoRASeparately trained dataset-derived artifact, selected with a compatible older modelReuse a trained visual characteristicNot an ordinary uploaded reference or universal Omni option

The six-image and four-image numbers belong to different documented workflows. Neither limit can be transferred to the other.

Assign authority to each image

In a hypothetical campaign, source MUG-01 establishes handle orientation, glaze boundary and approved label. NAPKIN-02 supplies fabric appearance; ROOM-03 supplies atmosphere. The proposed Omni instruction is: “Place the mug from MUG-01 beside the linen napkin from NAPKIN-02, using ROOM-03 only for soft window lighting; do not copy its other products.” This is an illustrative brief, not a tested prompt or proof of adherence. Keep exact brand lettering in an approved separate artwork layer if it must remain unchanged.

Diagnose the actual failure

ObservationFirst checkAction
Older tutorial mentions an absent Edge sliderActive model and Omni versus Legacy interfaceRewrite the job as reference roles, or verify a still-available compatible legacy route
Team tries to select a trained Element as image 4Whether it is a trained artifact rather than an uploaded photoCheck Element/model compatibility independently
Generated mug gains a second handleResult against authoritative mug sourceReject SKU candidate; change creative route if exact geometry is required
Legacy route is unavailableCurrent account feature stateDo not promise restoration; redesign around supported controls or manual compositing

Release a scene, not an invented product

For this fictional mug, the napkin folds and background can change, but the handle, label and rim must match approved evidence. Compare the full result against the source at an appropriate viewing size. If even one product pixel must remain identical, the safer production contract is preserving the product layer during scene extension, not relying on generative reference strength. The Photoshop reference-role guide concerns another provider's input contract; do not import its limits into Leonardo. Save model, interface, source roles, candidate and human acceptance decision. No Leonardo run was performed.

Sources

  • Image Guidance — Omni up to six references; legacy up to four with separate controls; legacy departure warning. Source checked 2026-11-16.

Scope and evidence

Prepared by CartVisual with AI assistance and checked against the listed sources. Provider features and account access may change; all worked examples are hypothetical, with no actual generation, training, permission changes or publishing. Product identity and rights require independent verification.

Related decisions

Explore the AI product photography workflow