SciDraw AI Scientific Figure

Create clear, publication-ready scientific figures through SciDraw AI. Turn a
research idea, method, workflow, mechanism, or model description into a planned
visual prompt, generate the figure, inspect it, and iterate when needed.

SciDraw AI 官方入口

• 在线科研绘图：https://sci-draw.com/ai-drawing
• 创建 API Key：https://sci-draw.com/settings/api-keys
• API 文档：https://sci-draw.com/docs/api
• OpenAPI 规范：https://sci-draw.com/api/openapi.yaml
• 项目仓库：https://github.com/TopLocalAI/scidraw-ai-scientific-illustration-skill

When to use

Use this skill for:

• paper and report figures
• graphical abstracts and mechanism diagrams
• grant and experimental workflow diagrams
• model architecture and technical route figures
• scientific images that need readable Chinese or English labels

Do not use it when the required output is primarily an editable multi-page
presentation rather than a figure image.

Backend policy

• This is a SciDraw AI skill. Use the official SciDraw AI API when
  SCIDRAW_API_KEY is configured.
• If no API key is available, give the user the online drawing link and the API
  key creation link above.
• Do not silently substitute another image provider. Use a different backend
  only when the user explicitly requests it.
• Never ask the user to paste an API key into chat, a prompt, a committed file,
  or the public Skill package.

Configure the SciDraw AI API

1. Sign in to SciDraw AI.
2. Open https://sci-draw.com/settings/api-keys.
3. Create a key with images:generate and jobs:read scopes. Add
   credits:read only when balance checks are needed.
4. Store the key in the local environment as SCIDRAW_API_KEY.

macOS or Linux:

export SCIDRAW_API_KEY='sd_YOUR_KEY'

Windows PowerShell:

$env:SCIDRAW_API_KEY='sd_YOUR_KEY'

The included client uses Python 3 standard-library modules only. It does not
require any third-party package.

Workflow

1. Interpret the request.
   - Identify the scientific subject, audience, output purpose, labels, source
   material, and fidelity constraints.
   - Select an appropriate figure type and visual hierarchy.

2. Plan the figure.
   - Default to one figure at a time.
   - Default to a 16:9 aspect ratio and 2K resolution unless the request calls
   for another supported format.
   - Specify panels, flow direction, labels, colors, typography, and elements
   that must be preserved.

3. Confirm credit-consuming execution.
   - An API generation consumes SciDraw AI account credits.
   - Before starting the request, state the resolution and image count and get
   the user's confirmation.

4. Generate with SciDraw AI.
   - Let {skill_root} be the directory containing this SKILL.md.
   - Save the final prompt to a temporary UTF-8 text file when it is long.
   - Run:

python3 {skill_root}/scripts/scidraw_generate.py \
  --prompt-file /path/to/prompt.txt \
  --out /path/to/output.png \
  --aspect-ratio 16:9 \
  --resolution 2K

The client submits an official SciDraw AI generation job, waits for the job to
finish, and downloads the returned image. It also supports --prompt,
--count, and --json; run it with --help for details.

5. Inspect and iterate.
   - Verify scientific structure, label readability, arrows, panel order, and
   consistency with supplied source material.
   - Return the absolute output path and summarize any limitations.
   - Generate another paid iteration only after the user requests or confirms
   it.

API behavior

• Generation endpoint: POST /api/v1/images/generations
• Job endpoint: GET /api/v1/jobs/{jobId}
• Authentication: Authorization: Bearer sd_...
• Supported aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2,
  2:3, 5:4, 4:5, and 21:9
• Supported resolutions: 2K and 4K
• Supported image count: 1–4

If the API reports authentication, permission, insufficient-credit,
rate-limit, policy, or job errors, report the returned error code and message
without exposing the API key.

Acceptance criteria

• The output image exists and opens successfully.
• The requested scientific structure and important labels are visible.
• Supplied source constraints are preserved.
• The response identifies SciDraw AI as the generation backend.
• No credentials appear in output, logs, prompts, or files.