fantasy-character-visual-dna

FANTASY CHARACTER VISUAL DNA

Organize the character's identity, physical structure, clothing, and signature elements into a recognizable visual image that can be further developed. Default to cinematic realism; user-specified styles take priority.

0. First explain the positioning and delivery structure

This Skill is responsible for character design and base image construction. All 19 example images in the gallery were generated and refined by the author using MJ + Image2, and are not final images generated by the Skill alone. Do not package the examples as a one-click direct output effect, and do not fabricate the per-image generation process or parameters.

Complete delivery structure: generated character base image + accompanying MJ prompt. At the beginning of a new user's first complete character design, use a short prompt to let them see:Character design → character base image + MJ prompt → MJ visual refinement → Image2 correction and refinement. Subsequent local modifications do not repeatedly insert long explanations; strict output requirements such as "prompt only" take priority.

• A complete new character design by default constructs one base image that makes it easy to observe the character, and delivers an MJ prompt with the same settings; the actual image is generated by the image tool provided by the current environment.
• When only the prompt is requested, do not call the image tool; when only the image, a specified platform, or a specified language is requested, deliver according to the user's requirements.
• When there is no image tool, clearly state that "the character base image has not yet been generated," provide base image generation instructions and the MJ prompt, and do not label a piece of text as a generated image.
• The base image prioritizes resolving the recognizability of the character's identity, proportions, face, clothing, and signature elements; subsequent image style, presentation, and refinement are left to the MJ + Image2 workflow. Other tools may be substituted according to the user's choice, without rewriting the gallery's source.

1. First choose the current task

What the user wants	Execution method	Read as needed
A new character or mythological reimagining	Clarify identity and a small number of core recognition points, and deliver one main proposal as a character base image + MJ prompt	Design rules、Prompt compilation
Reverse-engineering from an image or local modification	Assign reference image purposes, preserve identity and unmodified areas	Reference and editing
MJ / Image2 / Banana prompt	Deliver only copyable text in the required language and platform	Prompt compilation
Three-view, head-and-shoulders, expression, or prop assets	Use the confirmed baseline and change only the variables allowed this time	Asset consistency
Series characters or group portrait	First establish character differences, then unify material craft and worldview	Design rules
Confirmed Cyber Journey to the West project	Load the corresponding character preset on top of the general rules	Cyber Journey to the West preset
Feedback revision or result check	Locate the specific problem and correct it with minimal scope	Quality and correction

Read only the references needed for this task. The rule files are the basis for execution;Example galleryare visual cases, not templates that all new characters must copy.

2. Preserve user intent

Within the constraints of the system and execution environment, adopt in order: the current explicit requirements and latest corrections, the confirmed character baseline, the current project settings, and this Skill's default values.

• New requirements only override the corresponding fields; do not rewrite other already confirmed content.
• Style references do not automatically become identity references; partial clothing references do not automatically replace the face, pose, and background.
• Deliver the complete character design according to Section 0; when the user only wants prompts, output text only and do not generate images without authorization.
• Do not claim to have operated Midjourney, Image2, Banana, or other platforms that cannot be accessed; providing MJ prompts does not equal having completed MJ generation.
• When the input is basically sufficient, directly complete one main proposal. Ask only when key identity, baseline version, or hard requirements conflict with each other and cannot be judged.
• Do not automatically expand into a complete worldview, multiple rounds of questionnaires, batch variants, or a full asset set.

3. Establish the visual DNA

From the input, determine identity and situation, species, body proportions, facial hair, clothing materials, core props, pose, camera background, and aspect ratio. Supplement only the information necessary for this task.

First retain 1-3 key recognition points, then develop a small number of innovations. Translate words such as "premium, mysterious, cyber, dark" into silhouette, structure, material, or action, and avoid having only adjectives.

During design, check:

1. Identity and species. A complete beast demon has compatible logic for the head, neck and shoulders, torso, and limbs, and cannot degenerate into an ordinary person wearing an animal nose. Realism does not equal humanization.
2. Independent silhouette. Do not design everyone as broad-shouldered, muscular, caped, with large belts and spiked armor; character differences should still be visible when reduced in size.
3. Face and personality. Distinguish new characters through bone structure, age, proportions, gaze, and subtle asymmetry, and avoid influencer faces, xianxia mobile game faces, and uniformly refined beauty faces. When maintaining a reference person, do not use this to swap faces.
4. Understandable clothing and props. Clearly describe layers, connections, thickness, stress, and use, and do not replace design with random jewelry.
5. Details have primary and secondary emphasis. Focus on the face, core props, and a small number of recognition areas, while keeping complete quiet surfaces in other positions. Clean means reducing noise and fragmented ornaments, not erasing materials.
6. Wear has a cause. Let signs of use occur at reasonable contact and stress points, and do not replace realism with mud, rust, and rags all over the body.
7. Culture and technology have boundaries. Add technological elements only when needed, and make them serve identity, ability, or function. Cyber Journey to the West is not the default for all subject matter; Western characters do not automatically gain prayer beads, scriptures, or heavenly armor.

When the user actively requests an ornate, exaggerated, illustration, or animation style, follow that direction, while continuing to maintain identity, structure, and information hierarchy, and do not use the default aesthetic as a reason for rejection.

4. Make the image serve the task

• Use head-and-shoulders or a bust for looking at the face, two-thirds body for looking at clothing, and full body for looking at silhouette and complete props.
• Light and shadow must support recognition, and smoke, effects, darkness, and background must not swallow the core information.
• A single character usually uses a low-information-density background, with 1-2 environmental hints being enough; narrative images can be expanded according to the plot.
• Actions have footing, center of gravity, grip, and clothing response; do not apply the same pose to all characters.
• The aspect ratio follows the user and confirmed project agreements. When unspecified, 3:4 can be used to display a single character, and 16:9 can be used to organize a three-view or expression sheet; these are suggestions.
• New character assets have no text by default. Lettering, red seals, logos, and information bars are added only when the user requests them, and are not automatically copied from example images.

5. Lock references and assets

Distinguish between explicit user requirements, visible facts in the image, and design inferences. Reverse engineering does not equal recovering the original prompt; areas not shown do not equal having been restored.

After selecting the baseline, record: identity, species and proportions, face shape and facial features, hairstyle and hair color, clothing layers and color scheme, which side ornaments are on, weapon structure and count, and parts that cannot be changed. Left and right are based on the character itself; when necessary, add the position in the frame.

The three-view shares proportions and scale, changing only the viewing angle; expression sheets change only reasonable muscle states; local edits modify only the target area. When proposing compatible designs for unseen back structures, mark them as inferences, and include them in the baseline only after confirmation.

When the user requires exact counts such as nine teeth, nine beads, or multiple heads, the prompt should arrange sufficient visibility and count them in the actual output. If obscured or unclear, it cannot be reported as a "complete match."

6. Compile and deliver

Organize the prompt in the order of identity → body → facial hair → clothing materials → markers → action → composition and lighting → necessary constraints; remove repetitive and vague words.

• Chinese and English prompts share the same settings; provide a single language or both languages as the user needs, and do not add weapons, colors, or identity without authorization.
• Keep the platform's main description separate from parameters. Specific version numbers, reference parameters, and weights are filled in only when support has been confirmed; when verification is impossible, deliver the main text and state that the parameters have not yet been confirmed.
• Editing instructions must clearly specify the base image, items to preserve, objects to modify, division of reference roles, and lighting and occlusion transitions, and must not simultaneously redo the entire character during a local modification.
• A complete character design delivery includes a brief positioning statement, the actually generated character base image, and a matching MJ prompt; both share the same identity, body type, face, hair, clothing, color scheme, and markers. If the base image deviates, correct it first or clearly state the items to be corrected, and do not let the prompt quietly become another character.
• When the user requests conciseness or only the prompt, omit the design analysis and follow the specified platform and language; expand the DNA and design rationale only when the user requests a complete plan.

7. Check and finish

Check the identity anchors, species structure, division of reference roles, character differences, prop relationships, text-image consistency, and output format against the current request.

When receiving feedback, turn the issue into specific visible actions, save what has been confirmed, and prioritize correcting the erroneous parts. After actually generating an image, check based on the image; for text-only tasks, only a text check can be reported, and image testing cannot be declared passed.

Deliver once the user's requirements are met, and do not automatically start a new batch. Retain accurate explanations of model limitations and unverified items, and avoid promises of absolute consistency or one-click reproduction.