PaperBanana CN

One run from input to figure

The recording follows a methodology-diagram run from submitted inputs to the completed result.

Before you start

You need Python 3.10-3.12, uv, and a desktop browser.

Task	Required model connections
Methodology diagram	VLM and image generation
Statistical plot	VLM only
Multi-panel composition and run browsing	None

Each connection specifies its own protocol, Base URL, API key, model name, and timeout. The VLM and image roles may use the same service or two different services.

Quick start

1. Launch Studio

uvx paperbanana-cn studio

Open http://127.0.0.1:7860. uvx runs the package in an isolated environment and does not modify Debian or Ubuntu's system Python.

2. Add the model connections

Open Settings → VLM connection, fill in the service fields, and select Save and use. Repeat under Image connection before generating a methodology diagram.

Editing a saved connection does not activate it. An empty API-key field keeps the stored key. Studio does not fill stored keys back into the browser.

The connection guide lists the supported protocols, credential storage rules, connection tests, and legacy mode.

3. Generate a figure

Open Methodology diagram, provide the method content and figure caption, then choose an aspect ratio, resolution, and output format.

The same task from the CLI:

paperbanana-cn generate \
  --input method.txt \
  --caption "Overview of the proposed architecture" \
  --aspect-ratio 16:9 \
  --resolution 2K \
  --format png

What PaperBanana-CN adds

Separate VLM and image connections

The two model roles have independent protocol, Base URL, API key, model, and timeout settings. Studio, CLI, and MCP resolve the same saved connections. Saved profiles contain credential references; API keys remain outside the repository and run metadata.

Official APIs, OpenAI-compatible services, and Gemini-compatible services are supported. Provider specifics stay in the adapters rather than the scientific workflow.

Chinese and English Studio

The Studio interface, help text, validation, progress messages, and errors are available in Chinese and English. Changing the interface language does not rewrite prompts, paper text, or labels inside the generated figure.

Aspect ratios and resolution

Supported aspect ratios:

1:1 · 4:3 · 3:2 · 5:4 · 16:9 · 21:9 · 4:5 · 3:4 · 2:3 · 9:16

Resolution tiers:

1K · 2K · 4K

Studio shows the request size or provider-native tier before generation. If an adapter cannot produce the selected combination, validation stops the request and reports the unsupported option.

Studio workflows

Area	Workflow	Model connections
Create	Methodology diagram	VLM and image
Create	Statistical plot	VLM
Improve	Continue a saved run	Depends on the saved run
Improve	Quality evaluation	VLM
Automate	Full-paper orchestration	VLM and image
Automate	Batch generation	Depends on the task type
Automate	Parameter sweep	VLM and image
Tools	Multi-panel composition	None
Tools	Run browser	None

CLI, MCP, Docker, and Colab

Entry point	Command or link
Studio	paperbanana-cn studio
CLI	paperbanana-cn generate --help
MCP server	paperbanana-cn mcp
GitHub Action	Action reference
Docker	ghcr.io/mituan-ai/paperbanana-cn:2.0.1
Colab	Quickstart notebook

{
  "mcpServers": {
    "paperbanana-cn": {
      "command": "uvx",
      "args": ["paperbanana-cn", "mcp"]
    }
  }
}

The server provides 11 tools and reads the same active connections as Studio and CLI. See the MCP guide for the tool list and arguments.

docker run --rm -p 7860:7860 \
  -v paperbanana-cn-config:/home/paperbanana/.config/paperbanana-cn \
  -v paperbanana-cn-data:/home/paperbanana/.local/share/paperbanana-cn \
  -v paperbanana-cn-outputs:/work/outputs \
  ghcr.io/mituan-ai/paperbanana-cn:2.0.1 \
  studio --host 0.0.0.0

Install the command in a uv-managed environment:

uv tool install paperbanana-cn
paperbanana-cn studio

Run the current source checkout:

git clone https://github.com/mituan-ai/PaperBanana-CN.git
cd PaperBanana-CN
uv sync
uv run paperbanana-cn studio

The default package includes Studio, MCP, PDF input, OpenAI-compatible services, and Gemini.

Optional adapter	Install
AWS Bedrock	uv tool install "paperbanana-cn[bedrock]"
Anthropic	uv tool install "paperbanana-cn[anthropic]"
LiteLLM	uv tool install "paperbanana-cn[litellm]"
All optional providers	uv tool install "paperbanana-cn[all-providers]"

For CI, use paperbanana-cn connections add --api-key-env ENV_VAR so the key is read from an environment variable instead of a command-line value.

V1 and upstream

V2 is maintained on main as the paperbanana-cn distribution, the paperbanana_cn Python module, and the paperbanana-cn command.

• Browse the frozen v1 branch
• Download the v1.0.0 release

The scientific figure-generation core is based on llmsresearch/paperbanana. PaperBanana-CN is an unofficial community implementation and is not affiliated with or endorsed by the upstream authors.

Community

PaperBanana-CN is maintained by mituan under the MIT License.

• Ask usage questions in Discussions.
• Report reproducible bugs in Issues.
• Report vulnerabilities through Private Vulnerability Reporting.
• Read CONTRIBUTING.md before opening a pull request.

git clone https://github.com/mituan-ai/PaperBanana-CN.git
cd PaperBanana-CN
uv sync --extra dev
uv run pytest tests/ -q
uv run ruff check paperbanana_cn/ mcp_server/ tests/ scripts/

Do not upload API keys, private relay URLs, unpublished papers, private datasets, local connection stores, or generated run directories.

Star history