davidai1024/artifact-template-creator-cover

Creator Cover - Character Collaboration Cover

Generates a cover showing "this person is doing this thing" and supports a matching publishing title. When the user requests a cover, generate the image; when only a title, title research, brainstorming, or modifications to this skill are requested, execute within that scope and do not automatically generate images. The user's new requirements take precedence over the following defaults.

Task Entry

• Titles/title modifications only: Read title-methods.md. Based on the content genre, select a suitable mechanism from Expression Atlas and original examples. If a full understanding of the research basis is needed, read title-research.md. No need to read portraits or call image generation.
• Designing a cover: First, use the title methods to find a "reason to watch" from the material. Based on the content genre, read 2–4 relevant title examples and compare different angles. Then, finalize the cover copy and the character's workflow. Do not reverse-engineer a title from the easiest action to draw. If the user only wants an image, the internal selection is sufficient. If the user has already determined the title, prioritize keeping it and do not automatically change it.
• Title and cover together: First, determine one core promise of the content. Then, let the publishing title, cover headline, character action, and real outcome collectively express it.

Reading References

• When creating a character cover, first read Portrait Usage Guide to check the original portrait provided or explicitly specified by the user for this task.
• For first-time use, read the eight mechanisms of Xiaohongshu Cover Research. Based on the topic, select 1–3 mechanisms or cases from Composition Methods.
• For Douyin topics or cross-platform revisions, read Douyin Supplementary Research. Cases provide original post links; textual analysis can be used independently.
• Separate composition references, character identity, and real product materials. Cases can suggest camera angles and action ideas; the character's face shape, glasses, brand logo, and platform UI are determined by the current materials and task.

First, Decide Why the Person Is Here

Obtain the topic, audience benefit, actual product, and existing materials from context. Continue when both the theme and portrait source are clear. If the user requests to appear as themselves but has not provided a usable portrait, ask once.

Internally form a brief shooting plan:

1. What does the audience want to gain? Express in one specific benefit sentence.
2. What is the person doing? Choose a specific action, e.g., bringing close, comparing, opening, wearing, observing, operating, dodging, showing a result.
3. What image proves this? The product, real output, change, or environment.
4. How are hands, eyes, body, and items connected? What triggers the character's reaction?
5. Why is the clothing suitable for this theme? Work environment, usage method, season, or color scheme. Ordinary clothes are fine; no need to dress up for a role every time.
6. Camera distance, perspective, occlusion, and hierarchy: The face and result should both be visible; adjust proportions by theme.
7. One main headline, one necessary supplementary line, and a primary color derived from the theme.

You can internally compare several plans and choose the most expressive version; do not default to making the user do the creative selection. Check: "If the thematic item is replaced, is this action still completely universal?" If yes, prioritize strengthening the interaction. Ordinary holding/showing an item is also acceptable, but the item's direction, hand position, and gaze should serve its function.

Composition & Copy

• Distinguish between the publishing title and the cover headline. The former conveys the content object, highlight, or benefit; the latter highlights the key point readable on the thumbnail. They can repeat keywords; they are not forced to be different word for word. See title-methods.md for details.
• Decide on the narrative scene and text layout separately: First, choose a relationship like "operation scene, showing results, comparison, experiential reaction," then arrange the platform text area and cropping. Keep an action if it works; if the vertical crop makes it incomprehensible, change the angle or action.
• The person can be performing an action or appear in a product screen, photo, or scale comparison. First, ensure the theme and identity are clear; don't default to a half-body stance next to the product.
• Questions at the top and short words in the image are an optional structure. They should provide different information; if the user prefers minimal information, keep only one text focal point. The title bar should not squeeze out the face, hands, or result.
• Xiaohongshu covers default to a 3:4 vertical format. Adjust based on the user's given aspect ratio or publishing location.
• The person and thematic result should be seen at a glance, then read the title. The person is not fixed on the right side, not fixed to a half-body shot, and definitely not fixed to the original portrait's sitting posture.
• One cover focuses on one action, one main prop (or a related prop group), and one core benefit. Auxiliary details only help understand the main scene.
• Clothes and expressions can vary with the theme. Preserve the face shape, facial features, and recognizable accessories from the original portrait. Do not make a calm expression from a reference photo a permanent limitation, and don't make every expression an open-mouthed shock.
• Maintain hand-prop contact, realistic occlusion, shared lighting direction, and perspective. Products can be moderately enlarged for visual focus, but hands should not be hovering, and objects should not float without a narrative reason.
• Color scheme comes from the theme and items: Don't force black-yellow, black-red, or default purple gradients. The background should have a shooting or usage context; avoid unrelated light effects, floating cards, and decorative labels.
• Arrows, outlines, question marks, and partial enlargements are used only to indicate details or issues. The user prefers sophistication, minimal information, and direct results.
• Write copy that conveys user benefits. E.g., "Convert images to perler bead patterns" or "Generate a shopping list with beads". Don't use generic replacements like "destiny" or "unlock infinite possibilities". Do not arbitrarily promise free, full compatibility, precise time, or exaggerated effects.
• Color codes, quantities, specifications, parameters, prices, and product UI come from real materials. If data is missing, simplify the scene; do not fabricate a list that could be used for purchase. The cover illustration is not a deliverable high-definition blueprint or a real application screenshot.

Generation & Review

Use available imagegen image generation/editing tools and follow their current instructions. Do not use HTML, SVG, or programmatic drawing as a substitute for the real person cover requested by the user.

Clearly delineate input materials: The original portrait is only responsible for identity; real product/output for content accuracy; cases only for composition ideas. Prioritize using the original portrait and real product materials, supplemented by textual descriptions of the layout. If including case images from others, explicitly forbid copying the person's identity from them.

Generation instructions should include: purpose and aspect ratio, identity features of the person, specific action, relationship between the item, hands, and gaze, clothing and scene, camera and lighting, accurate Chinese text, and requirement for real data. Do not just write "Movie-like style".

After generating, check:

• Are the face shape, hairstyle, facial feature relationships, and accessories consistent with the creator's original portrait?
• Without reading the title, can you tell what the person is doing? Is the prop large enough? Is the result clear?
• Are hands, contact points, perspective, occlusion, and lighting natural?
• When shrunk to mobile feed size, is the main headline and subject clear? Is there too much text?
• Are there any typos in the Chinese text? Have real parameters been corrupted by the image generation?

If there are obvious issues, refine the problematic part; continue to use the original portrait when editing images to avoid using a chain of generated images as the sole source of identity. If tool failure prevents completion, honestly explain the situation; do not treat prompts as generated images.

Save the final image to the current workspace output directory and display it with a brief explanation. Only output multiple versions if the user requests multiple directions; do not default to generating an extra set of detail pages. Do not automatically publish to the platform.

Research Boundaries

Methods are derived from 495 Xiaohongshu posts and 551 Douyin works visible on November 8, 2026, including visible covers and page text. Coverage scope, cross-topic comparison, and observational boundaries are detailed in Xiaohongshu Cover Research, Douyin Supplementary Research, and Title Research. Original posts may change. Samples are used to summarize expression and composition; click-through and conversion effects need to be verified through published data.