Tutorial

How to Make a Comic with AI — What the Workflow Actually Looks Like

A real walkthrough based on what AI comic tools actually do today. No marketing fluff — just the actual steps, what works, and what's still being built.

June 16, 2026·12 min read
1

Write Your Script

Write in standard screenplay format — scene headings (INT./EXT.), character names, dialogue, and action descriptions. The parser looks for @-tagged characters like @Dante or @Vex and extracts them automatically. This is your blueprint. A good script with clear panel-by-panel descriptions gets dramatically better results than vague prompts.

Write panel descriptions the way you'd brief an artist: 'Medium shot, low angle — Dante grips the railing, rain streaming down his face, city lights blurring behind him.' Specificity pays off.
2

Lock Your Characters

For each character, define their traits: eye color, hairstyle, build, skin tone, outfit, scars, accessories. Upload 1-3 reference images. Set negative constraints — things the model should never do with this character. The system stores all of this in a database and retrieves it by name when you use @-tags in your prompts.

Reference images work best with consistent lighting and plain backgrounds. The system currently uses FLUX Kontext for single-character consistency and Nano Banana (Gemini 2.5 Flash) for multi-character scenes — both via Fal.ai. This is prompt-based and reference-image-driven, not a trained LoRA per character.
3

Pick Your Styles

Choose a House Style — this sets the overall medium and finish. Heavy-Ink Riso gives you that indie print look with visible paper grain and ink bleed. Neon-Noir for hot-pink synthwave. Inkstone Manga for B&W screentone. There are 13 house styles, each with its own prompt DNA and post-processing filter chain (style deepening, color toning, manga texture, grain). Then layer a Mood Style on top for individual scenes — Noir for tense interrogation rooms, Kinetic for action sequences.

Mix styles across chapters for visual contrast. Heavy-Ink Riso for the main narrative, switch to Datamosh glitch-art for flashback sequences. The visual shift signals genre changes.
4

Parse Your Script into Panels

Paste your screenplay-format script into the story parser. It extracts scenes, dialogue, character appearances, and action descriptions automatically. The engine assigns shot types (establishing, full, medium, close-up, extreme close-up) and computes page layouts for your chosen format — webtoon (vertical scroll), western (grid pages), or manga (right-to-left flow). This is a rules-based parser, not an LLM — it reads screenplay format and structures it into panel data.

Review the auto-assigned shot types. The parser makes reasonable guesses, but you know your story better. Swap a medium shot for a dramatic close-up when the moment demands it.
5

Generate and Iterate

Send panels to the generation API. When FAL_KEY is configured, images are generated via Fal.ai (Flux Dev/Pro, SDXL, or SD3 depending on the style). When a panel doesn't land — wrong expression, weird hand, off lighting — use the inpainting tool: draw a mask over the problem area, describe what you want instead, and Flux Pro Fill regenerates just that region. You can also regenerate entire panels that missed the mark.

Generate in batches of 3-4 panels, review, iterate, then move on. Don't get stuck perfecting panel 1 while the rest of the page waits. The enrichment step (optional Claude Haiku integration) helps by expanding vague references into concrete visual detail.
6

Export and Finish

The parser outputs structured panel data — coordinates, dimensions, shot types, and dialogue — as JSON. Currently, print-ready CMYK/PSD/PDF export is on the roadmap (Phase 2) but not yet built. For now, you can export generated images individually or use the parsed panel data to assemble pages in your preferred compositing tool.

The webtoon layout engine computes a continuous vertical scroll with auto-spacing. Western and manga formats compute per-page grids with proper gutters and margins. The layout math is solid — it's the final export format that's still being built.
Try It Yourself →