scientific-schematics – skill Dữ liệu & Khoa học cho agent AI

Scientific Schematics and Diagrams

Overview

Scientific schematics and diagrams transform complex concepts into clear visual representations for publication. This skill uses Nano Banana 2 AI for diagram generation with Gemini 3.7 Flash quality review.

How it works:

• Describe your diagram in natural language
• Nano Banana 2 generates a PNG draft from the specified components and relationships
• Gemini 3.7 Flash reviews quality against document-type thresholds
• Smart iteration: Refines when the score or verdict requests improvement
• Saved drafts and a review log for manual verification
• No coding, templates, or manual drawing required

Local review thresholds by document type (heuristics chosen by this helper, not publisher acceptance criteria):

| Document Type | Threshold | Description |

|---------------|-----------|-------------|

| journal | 8.5/10 | Nature, Science, peer-reviewed journals |

| conference | 8.0/10 | Conference papers |

| thesis | 8.0/10 | Dissertations, theses |

| grant | 8.0/10 | Grant proposals |

| preprint | 7.5/10 | arXiv, bioRxiv, etc. |

| report | 7.5/10 | Technical reports |

| poster | 7.0/10 | Academic posters |

| presentation | 6.5/10 | Slides, talks |

| default | 7.5/10 | General purpose |

Save diagrams under figures/ using -o figures/name.png, then inspect every label and relationship before including them in a paper or poster. The output location is the path you choose.

What the output is: a raster PNG at whatever resolution the image model returns. This skill has

no vector path and no DPI control — if a journal demands PDF, EPS, or 300 dpi TIFF, convert the PNG

downstream and check the result at final print size.

Quick Start: Generate Any Diagram

Run from this skill directory with requests installed and OPENROUTER_API_KEY set. The following paid-generation examples are illustrative; the request contracts are tested offline. Supply source-backed labels and relationships rather than asking the image model to invent them:

# Generate for journal paper (highest quality threshold: 8.5/10)
python scripts/generate_schematic.py "CONSORT participant flow diagram with 500 screened, 150 excluded, 350 randomized" -o figures/consort.png --doc-type journal

# Generate for presentation (lower threshold: 6.5/10 - faster)
python scripts/generate_schematic.py "Transformer encoder-decoder architecture showing multi-head attention" -o figures/transformer.png --doc-type presentation

# Generate for poster (moderate threshold: 7.0/10)
python scripts/generate_schematic.py "MAPK signaling pathway from EGFR to gene transcription" -o figures/mapk_pathway.png --doc-type poster

# Custom max iterations (max 2)
python scripts/generate_schematic.py "Complex circuit diagram with op-amp, resistors, and capacitors" -o figures/circuit.png --iterations 2 --doc-type journal

What happens behind the scenes:

1. Generation 1: Nano Banana 2 creates initial image following scientific diagram best practices

2. Review 1: Gemini 3.7 Flash evaluates quality against document-type threshold

3. Decision: If quality >= threshold and no improvement verdict → DONE (no more iterations needed!)

4. If below threshold or improvement requested: Improved prompt based on critique, regenerate

5. Repeat: Until quality meets threshold OR max iterations reached

Smart Iteration Benefits:

• ✅ Saves API calls if first generation is good enough
• ✅ Higher quality standards for journal papers
• ✅ Faster turnaround for presentations/posters
• ✅ Appropriate quality for each use case

Output: Versioned images (name_v1.png, name_v2.png), a copy of the latest successful version at the path you

asked for, and name_review_log.json with the score, critique, and any early-stop reason.

When the review cannot run — a rate limit, a content filter, a reviewer that answers in some

unexpected shape — the image is still generated and saved, but no score is invented for it. The log

records "score": null with the reason in "review_error". "reviewed": false means the reviewer did not return a usable response; a prose answer with no parsable score can have "reviewed": true. "final_reviewed" remains false without a numeric score, and the run

prints Review unavailable - image kept, quality not verified. Treat that image as unchecked and

look at it yourself and diagnose the review error before spending on another generation. A generation failure ends the loop without automatic replay; if an earlier draft exists, it is retained with its score and the failure in the log.

Configuration

Set your OpenRouter API key:

export OPENROUTER_API_KEY='your_api_key_here'

Get an API key at: https://openrouter.ai/keys

Data leaves the machine. Your prompt is sent to OpenRouter to generate the image, and the

generated image is sent back to OpenRouter for the quality review. Both are subject to OpenRouter's

data policies and those of the underlying model providers. Do not describe unpublished data,

patient information, or anything under embargo in the prompt.

Current API contract

The image call uses OpenRouter POST /api/v1/images with model, prompt, and n: 1; it reads data[0].b64_json and checks the returned MIME type (when supplied) and PNG signature. Quality review separately uses POST /api/v1/chat/completions with text and an image_url data URL. Both use Bearer authentication. The current chat schema still allows image output; this helper follows the dedicated image-generation guide without assuming the chat route was retired.

Models are google/gemini-3.1-flash-image (Nano Banana 2) and google/gemini-3.7-flash (review). Their IDs and modalities were checked in public catalogs; this helper's Gemini generation/review calls were validated offline, not with a paid end-to-end run. See [the verified contract and sources](references/iterative_refinement.md#openrouter-contract-reviewed-2026-09-30) before changing models or request fields.

For trial flows, use the [CONSORT 2025 template and item 22a](https://www.consort-spirit.org/item-22a-randomized), reconcile enrollment/allocation/follow-up/analysis counts, and report the specified primary outcome. For reviews, choose the appropriate [PRISMA 2020 template](https://www.prisma-statement.org/prisma-2020-flow-diagram); records, reports, and studies are distinct units. An AI score checks neither accounting system.

AI Generation Best Practices

Effective Prompts for Scientific Diagrams:

✓ Good prompts (specific, detailed):

• "CONSORT flowchart showing participant flow from screening (n=500) through randomization to final analysis"
• "Transformer neural network architecture with encoder stack on left, decoder stack on right, showing multi-head attention and cross-attention connections"
• "Simplified EGFR → GRB2/SOS → RAS-GTP → RAF → MEK → ERK pathway; label nucleotide exchange and phosphorylation distinctly"
• "Block diagram of IoT system: sensors → microcontroller → WiFi module → cloud server → mobile app"

✗ Avoid vague prompts:

• "Make a flowchart" (too generic)
• "Neural network" (which type? what components?)
• "Pathway diagram" (which pathway? what molecules?)

Key elements to include:

• Type: Flowchart, architecture diagram, pathway, circuit, etc.
• Components: Specific elements to include
• Flow/Direction: How elements connect (left-to-right, top-to-bottom)
• Labels: Key annotations or text to include
• Style: Any specific visual requirements

Scientific quality instructions (requested in the prompt; verify the result):

• Clean white/light background
• High contrast for readability
• Clear, readable labels (minimum 10pt)
• Professional typography (sans-serif fonts)
• Colorblind-friendly colors (Okabe-Ito palette)
• Proper spacing to prevent crowding
• Scale bars, legends, axes where appropriate

When to Use This Skill

This skill should be used when:

• Creating neural network architecture diagrams (Transformers, CNNs, RNNs, etc.)
• Illustrating system architectures and data flow diagrams
• Drawing methodology flowcharts for study design (CONSORT, PRISMA)
• Visualizing algorithm workflows and processing pipelines
• Creating circuit diagrams and electrical schematics
• Depicting biological pathways and molecular interactions
• Generating network topologies and hierarchical structures
• Illustrating conceptual frameworks and theoretical models
• Designing block diagrams for technical papers

How to Use This Skill

Simply describe your diagram in natural language. Nano Banana 2 generates it automatically:

python scripts/generate_schematic.py "your diagram description" -o output.png

The AI attempts:

• ✓ Layout and composition
• ✓ Labels and annotations
• ✓ Colors and styling
• ✓ Quality review and refinement
• ✓ PNG output for visual and scientific verification

Works for all diagram types:

• Flowcharts (CONSORT, PRISMA

… (nội dung đầy đủ có trong gói tải về)