Testing Leonardo for Character Continuity in a Comic
On this page
Editorial review and evidence boundary: This is a proposed comic workflow, not a verified studio adoption study. Flowith did not produce the 100-page series or measure the timing and productivity figures previously shown here.
Quick answer
Leonardo can be tested for concept art and character references, but a controlled panel set is needed before claiming consistent long-form output.
The earlier version asserted studio adoption, setup and training times, pages-per-day productivity, 100-page economics, and a named “Consistent Character Engine” workflow without sufficient first-party evidence. Those claims have been removed.
Start with a rights-cleared character sheet
Create a fictional character owned by the test team. Define:
- front, profile, and three-quarter views;
- face, body proportions, and age;
- hair, wardrobe, props, and markings;
- color palette and line treatment;
- allowed expression and pose range;
- details that must never change.
Do not train on an artist’s style or copyrighted character without permission.
Build a 24-panel continuity test
Use six scene types with four panels each:
- neutral dialogue;
- strong expression;
- full-body action;
- close hand-prop interaction;
- lighting and location change;
- two-character composition.
Record the current Leonardo product mode, model, references, prompt, seed if available, dimensions, attempts, and any custom training configuration.
Review every panel
Hide product labels and ask at least two reviewers to inspect:
- face and body identity;
- costume and prop continuity;
- line, color, and rendering style;
- spatial continuity;
- anatomy and contact;
- emotional readability;
- collateral changes after revision;
- lettering space and panel composition.
Keep all rejected generations. Report accepted panels per total attempts and manual correction minutes; do not publish a pages-per-day number from a short test.
Separate generation from comic production
A complete comic still requires script, thumbnails, pacing, panel layout, backgrounds, lettering, balloon order, accessibility, color management, prepress or digital export, editorial review, and rights clearance.
Test generated material as one production input, not a replacement for the whole workflow.
Customization and LoRA
If the current product offers model training or reference controls, verify dataset requirements, base compatibility, settings, output rights, deletion, and whether the model can be exported. Use consented, licensed data and preserve a dataset inventory.
Do not quote a universal training time. Record the exact configuration and result.
Production gate
| Gate | Pass evidence |
|---|---|
| Character | Defined traits persist |
| Style | Variance within tolerance |
| Story | Panel action is readable |
| Corrections | Editable within budget |
| Rights | Inputs and outputs cleared |
| Disclosure | Publisher policy satisfied |
| Export | Required print/digital format |
| Ownership | Source and final assets archived |
Only expand beyond the pilot after these gates pass. Human artists and editors remain accountable for the published work.