Docs/Features

StyleDNA

StyleDNA is Nephele's style-personality analysis tool. It doesn't grade your work or judge whether it's good or bad — instead, like MBTI, it analyzes your visual DNA, locating your stylistic leanings across 5 dimensions and generating a 5-letter code (such as LSXUA) to help you see the "personality" of your own work more clearly.

技巧

StyleDNA deducts stamina based on actual token usage — typically about 5–15 points per analysis, lower than AI Critique. Generating a share card costs no extra stamina.


Where to Find It

Entry pointHow to open
ToolboxLeft navigation bar → Toolbox → Page 2 → StyleDNA
Agent chatJust tell Nephele "analyze this piece's style" or "generate its style DNA"
Drag and dropDrag an image into the Nephele window and the Agent can run the analysis automatically

How to Use It

  1. Upload your work — supports PNG / JPG / WebP / BMP / TIFF
  2. Click "Start Analysis" — analysis usually takes 5–10 seconds
  3. Review the results — 5-letter code, 5 dimension sliders, technique analysis, style lineage, reference colors, and tags
  4. Generate a share card — click "Save Card" or "Copy Card" to share it on social platforms

提示

Once the analysis is done, you can generate a polished 1080×1350 share image and post it straight to Instagram, Xiaohongshu, or Twitter.


Reading the Results

The 5-Letter Code

The heart of StyleDNA is a code made of 5 letters, such as LSXUA. Each letter represents a leaning on one dimension:

DimensionLeft poleRight poleCoding rule
Line vs. ColorL (line-driven)P (color-driven)< 50 takes the left, ≥ 50 takes the right
Realistic vs. StylizedR (realistic)S (stylized)< 50 takes the left, ≥ 50 takes the right
Minimal vs. ElaborateM (minimal)X (elaborate)< 50 takes the left, ≥ 50 takes the right
Vivid vs. MutedV (vivid)U (muted)< 50 takes the left, ≥ 50 takes the right
Narrative vs. AtmosphericN (narrative)A (atmospheric)< 50 takes the left, ≥ 50 takes the right

For example, LSXUA means: line-leaning, stylized, elaborate, muted, and atmospheric.

32 Style Types

5 dimensions × 2 directions = 2⁵ = 32 base style personalities. Even works that share the same code differ subtly — the exact score (0–100) determines your precise position on each spectrum.

Technique Analysis

A smooth, natural-language analysis covering your work's:

  • Line style — hard-edged inking, soft-edged rendering, lineless, and so on
  • Coloring approach — flat color, painterly, cel-shading, watercolor texture, and the like
  • Light and shadow — soft diffusion, strong contrast, atmospheric lighting, etc.

Style Lineage

An analysis of which traditions your style inherits from and which periods or regional styles have influenced it (for example, "Japanese cel-shading tradition + 2010s digital grain-texture trend").

Style Description

A natural-language description, written as if "introducing this artist to a friend," that captures the visual temperament of your work.

Reference Colors

Six dominant colors extracted from your work, ranked by frequency. You can use these values as a color reference.

Style Tags

3–5 English style tags (such as cel-shading, anime, grain-texture) that make it easy to search for similar work on social platforms or asset sites.


Share Card

Once the analysis is complete, you can generate a polished style profile card:

  • Size: 1080 × 1350 (4:5, well suited to Instagram/Xiaohongshu)
  • Layout: work thumbnail + large 5-letter code + 5 dimension slider visualization + reference colors + technique/lineage/description + style tags
  • Design: an editorial, minimalist white card, with each letter rendered in its dimension's color

Click "Save Card" to generate a PNG locally, or click "Copy Card" to paste it directly into a chat window.


How It Differs from AI Critique

AI CritiqueStyleDNA
Core question"Is this piece good?""What personality does this piece have?"
MethodMulti-dimension scoring + 10-tier rating5-dimension bipolar mapping + 5-letter code
Good or bad?Yes (higher/lower scores, better/worse ratings)No (both 0 and 100 are valid styles)
Output focusScores, improvement suggestions, learning pathCode, technique, lineage, description, palette
AnalogyAn art-exam grading teacherAn art historian + an MBTI personality test
Social appealLowHigh (share cards)
Stamina costAbout 10–20 pointsAbout 5–15 points

技巧

We suggest pairing the two tools: first use StyleDNA to understand where your style sits, then use AI Critique to find the concrete room for improvement within that style.


Using It in Agent Chat

In an Agent chat, you can invoke StyleDNA like this:

"Analyze this piece's style"

"I've drawn a series of works on the same theme — help me check whether their styles are consistent"

"Analyze this image's style, then write me a caption suitable for posting on Xiaohongshu"

The Agent can combine style_code, description, and style_tags to write social-media captions.


FAQ

Q: Will the style code change?

Almost never. The code is computed locally and deterministically — for each dimension, as long as the score crosses the 50-point midline it takes the corresponding letter, otherwise the other end. So small fluctuations (like 45→48) usually won't change a letter; only swings that cross the midline change the code, which makes the 5-letter result highly stable.

Q: Can I analyze someone else's work?

Technically yes, but we strongly recommend analyzing only your own work — the whole point of StyleDNA is to understand the visual personality of your own pieces. Please respect other creators' rights.

Q: How are the reference colors extracted?

Using a classic color-quantization algorithm from computer graphics, it extracts the 6 most frequently occurring colors from the image. This is purely local computation and costs no stamina.

Q: What if the fonts on the card don't render correctly?

The card generator prefers the software's bundled fonts and automatically falls back to a system font (such as Microsoft YaHei) if loading fails, ensuring it renders correctly on every system.


AI-generated content is for reference only. There's no good or bad in style — every form of visual expression has its own unique value.

Last updated Jun 21, 2026·Applies to v0.5.2-beta