CLI Reference

CLI Reference

PaperBanana provides four CLI commands: generate, plot, evaluate, and setup.

paperbanana generate

Generate a methodology diagram from text.

paperbanana generate \
  --input method.txt \
  --caption "Overview of our framework" \
  --output diagram.png \
  --iterations 3

Flag	Short	Description	Default
--input	-i	Path to methodology text file	Required
--caption	-c	Figure caption / communicative intent	Required
--output	-o	Output image path	Auto-generated in outputs/
--iterations	-n	Visualizer-Critic refinement rounds	3
--vlm-provider		VLM provider name	gemini
--vlm-model		VLM model name	gemini-2.0-flash
--image-provider		Image generation provider	google_imagen
--image-model		Image generation model	gemini-3-pro-image-preview
--config		Path to YAML config file	configs/config.yaml

Output structure:

outputs/run_<timestamp>/
├── final_output.png      # Final generated diagram
├── iteration_1.png       # After first refinement
├── iteration_2.png       # After second refinement
├── iteration_3.png       # After third refinement (same as final)
├── planning.json         # Retrieved examples and descriptions
└── metadata.json         # Run config, timing, provider details

paperbanana plot

Generate a statistical plot from data.

paperbanana plot \
  --data results.csv \
  --intent "Bar chart comparing model accuracy across benchmarks"

Flag	Short	Description	Default
--data	-d	Path to data file (CSV or JSON)	Required
--intent		Communicative intent for the plot	Required
--output	-o	Output image path	Auto-generated in outputs/
--iterations	-n	Refinement iterations	3

The plot pipeline generates Matplotlib code via Gemini rather than using image generation directly. The generated .py file is saved alongside the output image for inspection.

paperbanana evaluate

Compare a generated diagram against a human reference using VLM-as-a-Judge.

paperbanana evaluate \
  --generated diagram.png \
  --reference human_diagram.png \
  --context method.txt \
  --caption "Overview of our framework"

Flag	Short	Description	Default
--generated	-g	Path to generated image	Required
--reference	-r	Path to human reference image	Required
--context		Path to source methodology text	Required
--caption	-c	Figure caption	Required

Outputs scores on four dimensions with hierarchical aggregation:

• Primary: Faithfulness, Readability
• Secondary: Conciseness, Aesthetics

See Evaluation for details on the scoring methodology.

paperbanana setup

Interactive first-time configuration wizard.

paperbanana setup

Opens your browser to Google AI Studio for API key creation and saves it to .env. See Installation for manual setup.

Global Flags

These work with any command:

Flag	Description
--help	Show help for the command
--verbose	Enable debug logging