jinchenma-ip-article-illustrations

Jinchenma IP Article Illustrations

Before Starting

Read the following files completely:

• references/ip-pack-format.md: Initialize the article character configuration, import, activate, switch, and parse the current IP.
• references/illustration-style.md: Apply the shared jinchenma-surrealism visual specifications.
• references/article-workflow.md: Determine the number of images, cognitive anchors, generation, saving, and delivery methods.

Initialize and Display the Current IP

Before analyzing the article, follow references/ip-pack-format.md to complete character parsing:

1. When the user requests to import or switch characters, complete the corresponding process first.
2. On the first run and when there is no article character configuration, in an interactive scenario first prompt:当前默认使用 Jinchenma。是否要更换为你自己的 IP 形象？ When the user chooses to change, enter the character import process; when the user chooses not to change, use the built-in Jinchenma.
3. Custom characters are read only from the article Skill's own ips/ . Builder packages and external files are read and copied only during explicit import.
4. Parse and load only one character IP's manifest.json, three-view images, and character specifications.
5. Before starting image generation, use an intermediate process prompt:当前使用 IP：<displayName> (<ip-id>)。

The current IP is a character identity, not the article's art style. All character IPs continue to use jinchenma-surrealism。

Execute the Workflow

1. Read the complete article or all text provided by the user. When the user provides an article address, obtain the complete body text, and do not infer content solely from the title.
2. Extract the core viewpoints, key turning points, article emotions, and cognitive anchors truly worth visualizing.
3. Determine the number of images based on the body text length and information density, merge duplicate content, and do not cover content evenly by chapter.
4. For each image, determine one core judgment, one surreal relationship, and 1–3 main objects.
5. Arrange the current IP's narrative role, character scale, actions, and emotions suitable for the current content.
6. List a small amount of short text that must be presented accurately; the text language follows the input article.
7. Complete the image plan internally, do not output the plan first, and do not wait for secondary confirmation.
8. Use the current character IP's three-view images as multi-angle references for the same character, generate each image independently, and do not replace them with a collage.
9. Check the article meaning, text accuracy, character consistency, and image density one by one; correct only one main issue at a time.
10. Save images according to the output rules. Keep the final reply brief, listing each image, purpose, absolute path, and suggested insertion position in order.

Character and Style Boundaries

• The shared process reads character identity, clothing, accessories, and stable colors only from the current character IP.
• Switching characters only replaces the character identity; do not change article analysis, image count, cognitive anchors, metaphor design, composition, saving, and delivery rules.
• Understand the three-view images as the front, side, and back of the same character, and do not generate three characters.
• Let the character participate in the core relationship, but do not let the character overpower the article's viewpoint.
• Do not use original real-person photos to replace the three-view images within the character IP.
• Do not write accidental style drift from a single generation back into the character IP.

Output Boundaries

• Article illustrations are preferentially saved to the current project root directory .jinchenma-assets/<article-slug>/。
• When there is no project, the project is not writable, or the host sandbox prohibits writing, fall back to ~/.jinchenma-assets/<article-slug>/。
• Do not overwrite existing images; use -v2、-v3 incrementing version.
• .jinchenma-assets/ Save only article illustrations, not three-view drawings, character specifications, or private IP.
• By default, treat the input article as read-only, and only briefly provide the insertion position of each image in the final reply.
• Only modify the original article when the user explicitly requests that image references be written into the article.