Based on PaperBanana's dev_planner
mode, i.e. Retrieve → Plan → Visualize.
The output image can be below expectation. You can tweak the image later with the "edit" block.
The dataset and prompts are from the original PaperBanana repo.
PaperBanana automates academic illustration generation for AI scientists. It transforms research content into publication-ready diagrams through an intelligent pipeline combining retrieval, planning, and visualization.
- Retriever: Searches curated diagram collections to find visually similar references for guiding illustration generation
- Planner: Translates research content into detailed visual prompts using in-context learning from reference diagrams
- Visualizer: Generates scientific illustrations from text prompts using state-of-the-art image generation models
- Editor: Modifies existing images based on text instructions using AI-powered editing capabilities
- Generate (Subflow): End-to-end academic illustration pipeline combining retrieve, plan, and visualize blocks
- Full Pipeline: Use Generate subflow for automated end-to-end illustration from research content
- Manual Pipeline: Chain Retriever → Planner → Visualizer for step-by-step control
- Iterative Refinement: Generate → Editor for fine-tuning outputs that need adjustments
- Reference-Based: Use Retriever output to guide Planner for consistent visual style
- Provide research content describing mechanisms, processes, or findings
- Optionally add a caption for additional context
- Run the Generate subflow or individual blocks in sequence
- Review and edit the generated illustration if needed
Example 1: Mechanism Diagram
- Input: "Research on thyroid medullary carcinoma mechanisms showing PGC-1α expression, PROS1 acetylation, and tumor-macrophage interaction via PROS1/MERTK"
- Output: Publication-ready mechanism diagram with normal vs. cancer cell comparison
Example 2: Process Flow
- Input: "Single-cell sequencing workflow from sample collection through clustering analysis to pathway enrichment"
- Output: Visual workflow diagram showing each analysis step
Example 3: Image Editing
- Input: Generated image URL + prompt "Add labels for key cellular components"
- Output: Updated diagram with annotated labels
MIT.