Creative Review
A creative review is no longer just "putting a number on it." It looks back at this piece from an illustrator's point of view: what you pulled off, where you can still explore, what to try in your next piece, and it maps out a learning path for the directions worth a long-term investment.
If you want the hard, numbers-driven view, it also keeps a "Commercial Evaluation" tab — letter grades, a dimension radar, and market price references all live there.
Two perspectives, one analysis — just switch tabs.
提示
Creative Review was formerly called "AI Evaluation" — both names are recognized in the toolbox card and in Agent commands.
技巧
AI Evaluation deducts stamina based on actual token consumption, usually about 10–20 points per run.
Where to Start
| Entry point | Action |
|---|---|
| Toolbox | Left nav bar → Toolbox → Page 1 → AI Evaluation / Creative Review |
| Agent chat | Tell Nephele "help me review this piece" (leans toward the growth view) or "evaluate this piece and grade it" (leans toward the commercial score) |
| Drag and drop | Drag an image into the Nephele window, and the Agent will recognize it and invoke the tool automatically |
How to Use
1. Upload Your Work
PNG / JPG / WebP are supported. We recommend uploading the original at a higher resolution so the AI can see it more clearly.
2. Fill in a Style Description (optional)
If you want the AI to evaluate against a specific stylistic standard, you can fill in a style description, such as "Japanese cel-shaded," "oil-painting realism," or "pixel art." If you leave it blank, the AI will automatically identify the style genre and judge by the standards of that style.
提示
The style description determines "what ruler the AI measures with." For example, judging a chibi piece by realistic standards for anatomy would be unfair — once you fill in the style description, the AI switches to the professional standards for that style.
3. Start
A single analysis usually takes 5–15 seconds, depending on your network and the complexity of the image.
Results: Two Perspectives
When it's done, the results are shown in two tabs, landing on "Growth Review" by default.
Growth Review (default)
A look-back from an illustrator's perspective, without piling on numbers:
- What you pulled off — exactly which parts of this piece work (pointing to specific elements in the image, not vague praise)
- What you can explore — from a review perspective, where you can still go deeper
- What to try next — a single-sentence, immediately actionable next step
- Directions worth a long-term investment — picks 2–3 dimensions and gives a full learning path (see the table below)
- Recommended artists / find similar artists — gives you one well-matched benchmark artist, with one click to go find more people like them
- Reverse trace — see how this piece is doing across platforms (views / likes / saves, etc.), about 30–90 seconds
The learning path under each "long-term direction" includes:
| Content | Purpose |
|---|---|
| Actionable suggestions | Specific adjustments you can make right away |
| Practice methods | Systematic day-to-day training |
| Reference artists | Benchmarks for that dimension (with Pixiv links) |
| Tutorial search keywords | Paste directly into Bilibili / YouTube |
| Reference-image search keywords | Paste directly into Pinterest / Pixiv |
Commercial Evaluation
For the hard, score-based view, switch to this tab:
- Overall grade — letter grade (S to E) + grade label + overall assessment
- Style detection — the detected style name and tags
- Dimension radar — dual radar charts of general dimensions + specialized dimensions, with an item-by-item diagnosis of each dimension (grade badge + score bar)
- Market price reference — for work that has a matching category (such as illustration), gives an indicative commission price range (personal / commercial) to give you a sense of the commercial scale of the piece
提示
The market price reference is drawn from publicly posted commission guide prices on external platforms and only conveys a sense of scale — it is not Nephele pricing or making any promise about your work. There is currently no matching category range for UI / Logo / photo retouching, etc.
10-Tier Grading System
The commercial evaluation uses a gamified 10-tier grading scheme to make skill levels more recognizable:
| Grade | Label | Rough meaning |
|---|---|---|
| S | Hall-of-fame level | Top of the industry, extremely rare |
| S- | Master level | Close to textbook caliber |
| A+ | Professional elite | Commercial benchmark level |
| A | Professional, excellent | A mature professional standard |
| B+ | Mature | Polished enough for commercial use |
| B | Competent | Solid fundamentals, with a clear style |
| C+ | Growing | Understands the principles, with gaps in execution |
| C | Has fundamentals | Basic concepts are established |
| D | Getting started | Fundamentals still need consolidation |
| E | Emerging | Just beginning to explore |
技巧
Don't compare Nephele's scores directly with other platforms. Its frame of reference is "the full spectrum of practitioners in that field worldwide," so the score may be lower than some "courtesy-grade" platforms — but that's precisely to let you see real room for growth. The Growth Review perspective deliberately doesn't give you a chance to fixate on the absolute number, for exactly this reason.
Type-Aware Evaluation
Nephele doesn't measure every piece by the same standard. Different types have completely different specialized dimensions.
General Dimensions (evaluated for every type)
- Composition / layout — visual guidance, spatial allocation, image balance
- Use of color — color harmony, contrast, color mood
- Ingenuity and expression — original ideas, differentiation, personal style
- Finish — refined details, edge handling, overall polish
- Overall tone — stylistic consistency, emotional delivery, thematic fit
Specialized Dimensions (assigned by type)
| Work type | Specialized dimensions |
|---|---|
| Illustration / concept art | Technique, light and shadow, form expression, spatial composition, detail |
| Chibi / SD | Form appeal, deformation consistency, expressiveness |
| UI / interface design | Information hierarchy, usability, visual conformity, adaptability |
| Logo / branding | Recognizability, scalability, brand fit, simplicity |
| Pixel art | Technique, light and shadow, form expression, spatial composition, detail |
| Photography / photo retouching | Technique, light and shadow, form expression, spatial composition, detail |
| Other visual work | Technique, light and shadow, form expression, spatial composition, detail |
You can specify the work type manually, or let the AI detect it automatically. Design categories like UI / Logo also switch to dedicated calibration anchors (referencing SaaS / design-system standards) and adjust the scale for differences like dark / light and minimal / data-dense.
Using It in Agent Chat
You can invoke it like this:
"Help me review this piece and figure out which direction to train for the next one"
"Evaluate this app interface — the type is UI design — and grade it"
"Look at this piece from a commercial illustration angle, focusing on its appeal as a product cover"
Saying "review" leans toward the growth perspective; saying "evaluate / grade" brings out the commercial score. The Agent turns your request into evaluation parameters and returns structured results.
FAQ
Q: Why is my score lower than I expected?
The commercial evaluation's frame of reference is the global spectrum of practitioners, so the standard runs strict. It's normal for most amateur work to land between 30 and 65. We recommend focusing on the grade and the specific suggestions in the Growth Review, rather than fixating on the absolute score.
Q: What if the AI gets my work type wrong?
Tell the Agent the type explicitly before evaluating, e.g. "this is a pixel-art piece"; or fill in a style description in the tool interface to guide it.
Q: If I evaluate the same image multiple times, will the score differ?
Almost never. The score shown at the end isn't the raw score the model gives directly; it's computed through fixed piecewise stretching and weighted recalculation, so results for the same image are essentially stable. Any change is usually due to a server-side model version update.
Q: Are my images saved?
No. Images are compressed and encrypted before being sent to the cloud AI for analysis, and the server does not store the originals.
Q: Are the search keywords in the learning path just templates?
No. Every suggestion is generated based on the weaknesses of your specific piece, including targeted practice methods, reference artists, and search keywords.
AI-generated content is for reference only; the final judgment rests with the creator.