Visual & data

Comic image sets

Twelve theme groups, three-IP consistency, nine-image prompt planning and batch QA.

Paid $9 list, $7 until 2026-12-31 · one skill, yours for good. The outline, the steps and the file open after purchase.

Background

Use this skill to plan, generate, review, and archive image sets for the <PROJECT_CODENAME> project. Preserve the locked twelve-group structure and keep the three main IP characters consistent across all 108 images.

Why it is built this way

The slot count is locked before anything else: 12 groups, 9 images each, 108 total, and the group names do not move. That freeze is what makes the style rules testable. A landmark such as 龙头山 and 汉山 is placed inside 秦岭以南 by replacing ordinary mountain frames, and 兴汉胜境 night scenes stay inside 江岸灯火, because growing the count to fit a new subject breaks every downstream count. With 108 fixed, a confusable pair of groups in a thumbnail grid becomes a defect you can see and fix.

Three to five candidates per image exist because the first output is never the final one. Character drift, extra limbs and text artifacts surface on generation and get corrected through inpainting only after selection is stable, then upscale and grade. The record step carries prompt, negative, model version, seed, LoRA, IP-Adapter and ControlNet so an approved frame can be reproduced, and readable text is added in post rather than asked for from the model.

Where it bites

• A group that outgrows its nine slots A new landmark gets its own extra frames or its own group, and the total drifts off 108, which breaks the images.json count and every later percentage of character presence. The structure section fixes landmarks as replacements inside their existing group.
• Two groups that read the same Two groups end up sharing a style family, and in a thumbnail grid they look like one series, so the album loses a chapter without anyone noticing. The contract forbids two groups sharing a primary style family and sends the pair back to the isolation matrix in style-bible.md.
• The character comes back as someone else The same character returns with a different face shape or wrong clothing because the reference tokens were left out of one prompt, and the fix is not a full regenerate. QA rejects drift, clothing errors and extra limbs, and correction happens through inpainting after a candidate is chosen.
• Readable characters baked into the frame A prompt asks for a shop sign and the model returns plausible-looking glyphs that turn out to be nonsense, which cannot be edited cleanly later. The guardrail keeps Chinese text, calligraphy, inscriptions and slogans out of generated frames, and type gets added in post-production.

When to use it

Produce and review coherent text-to-image sets for the <PROJECT_CODENAME> project, including twelve fixed theme groups, three-IP character consistency, nine-image prompt planning, batch generation, seed and model metadata, visual QA, and AI compliance. Use for this project's image production or review; not for general UI design or unrelated image generation.