MakeCorn vs Leonardo.Ai: Build a Character Reference Pack You Can Revise
A character pack needs agreement about what should stay recognizable. “The same person in five settings” leaves too much unstated: are you preserving the face, coat, body proportions, palette or all of them? A practical MakeCorn vs Leonardo.Ai evaluation starts with a written identity sheet and a small reference pack.
Our original example is Mara, a fictional thirty-six-year-old museum guide. She has short dark curls, round amber glasses and a moss-green jacket. The pack contains a neutral portrait, a full-length gallery view and an outdoor poster illustration. All scenes are ordinary, fully clothed and fictional. Public official sources were checked on 21 October 2026. We did not upload references, purchase plans or test image quality.
MakeCorn and Leonardo expose different reference workflows
MakeCorn’s editor describes selecting a source photo and using a written Custom edit. It asks users to compare the result with the original and warns that exact preservation is not guaranteed. This gives us a documented starting point for an image-led revision evaluation. MakeCorn image editor.
Leonardo’s Omni editor guide describes natural-language edits with references, including up to six images depending on the selected model. It allows references for different elements such as a character, setting or color scheme. That model dependency matters: do not assume every model accepts six references or handles them identically. Leonardo Omni models and Inline Editor.
| Pack requirement | Proposed evaluation |
|---|---|
| Preserve Mara’s face while changing the gallery | Begin with the approved portrait; inspect identity changes |
| Borrow a color palette without borrowing a face | Label the reference’s purpose explicitly |
| Combine a character and a new setting | Test only the references supported by the chosen model |
| Prepare a poster with exact text | Generate the illustration and typeset the headline separately |
These are proposed tests, not guarantees. A useful pack can contain a few controlled images rather than many loosely related generations.
Separate the identity sheet from the style board
The identity sheet records fixed features: age thirty-six, hair shape, glasses, jacket and broad proportions. The style board records optional choices: muted print texture, warm light and a restrained green-and-ochre palette. Keep those files separate so a later request for a different illustration style does not inadvertently change the character.
Do the same for a location reference. A photo of a gallery can supply layout without becoming the intended source of a person’s face or a copyrighted artwork. Use a location image you created or are authorized to use, with no unrelated identifiable person in it. An image being online is not evidence that it is a suitable reference.
For our first proposed revision, keep the neutral portrait and change only the background. A suitable original instruction is: “Place this fictional adult museum guide in a quiet sculpture gallery with plain stone plinths. Keep her short curls, amber glasses and moss-green jacket. Use soft daylight. Do not add text.” This tells the reviewer what to check as well as telling the generator what to attempt.
Evaluate a reference pack in three passes
First, inspect identity without changing the visual style. Compare the neutral portrait and gallery result side by side at the same face size. Check glasses shape, hairline, apparent adult age and jacket details. Record specific drift rather than assigning a universal likeness score.
Second, change framing. Ask for a full-length view in the same setting. Check whether proportions, clothing and accessories remain plausible. A face match in a close-up does not establish that the full-body version fits the character sheet.
Third, change presentation style. Create a poster illustration from the approved character, then compare whether the identity cues survive the new texture and palette. Leave headline space empty and add the final words in a conventional design tool.
This sequence helps locate a problem. If identity works until the style changes, adding more background descriptions is unlikely to address the right issue. Revert to the last approved result, reduce the style change, or choose which identity cues are essential for the poster.
Set privacy before making the pack
Leonardo’s private/public image help says private generation is available to paid members and controlled by the Private Mode toggle. Public images have additional visibility and usage implications. Check the state before generating, rather than relying on a later privacy change to undo earlier exposure. Leonardo private/public images.
MakeCorn says its generated content is private by default. That describes the default audience state, not a guarantee that files avoid hosted processing or storage. MakeCorn privacy policy.
For an unreleased fictional character, our recommendation is to decide in advance who may see the references, drafts and approved images. Use a project folder with clear names, and avoid real customer information in the brief. If contractual confidentiality is strict, confirm the provider arrangement directly before uploading.
Count edits and discarded variants as part of the pack
Suppose you need three approved views and allow four attempts for each. That is twelve attempts, before separately charged upscaling or further revisions. This is a chosen budget ceiling, not a measured success rate. Record each requested output count and the actual quote shown for the selected model.
Do not compare one platform’s token allowance with another’s credits as raw quantities. Instead record: full payment, included allowance for the same period, cost of each selected operation and the number of images finally accepted. If a current quote changes when you add references or request more outputs, update the whole pack estimate.
A revision log can be simple: version, source reference, one changed instruction, charged units, accepted/rejected and reason. “Glasses changed from round to rectangular” helps the next attempt. “Not good” does not. Retain approved originals so a later edit can start from a stable version instead of the most recent experiment.
Which workflow should you shortlist?
Shortlist MakeCorn when its source-photo plus written-edit route fits the pack and you prefer evaluating that simple handoff. Shortlist Leonardo when you want to test combining explicitly assigned reference roles through one of its supported Omni models. Neither public page establishes that one preserves Mara better; only a controlled evaluation of your own brief can answer that.
Do more references automatically improve consistency?
No. Additional references can introduce competing faces, clothes or composition. Give every reference a clear purpose and stay within the selected model’s support.
Can I use an actual person’s photo for this test?
This example uses an original fictional adult. If you use other material, obtain the required rights and consent first; do not infer permission from public availability.
What should I prepare before paying?
Write the identity sheet, make or lawfully obtain the reference files, set the intended privacy state and cap the attempt budget. Start with one background revision before ordering an entire pack.
Related decisions: Browse all five comparisons or use the homepage planner.