PaperBanana: Automating Academic Illustration for AI Scientists: plain AI paper summary
AI-generated summary. Read the original paper before relying on it.
Hugging Face paper page · Original paper on arXiv
PaperBanana is an agent workflow for making academic diagrams and plots. Agents retrieve references, plan the content and style, render an image, then critique and refine it.
Key findings
- The workflow combines vision-language and image-generation models.
- Reference retrieval supplies material for the planned figure.
- Separate planning and rendering steps turn the requested content into an illustration.
- An iterative critique step checks the rendered result and guides refinement.
How you might use it
- Use it for one internal technical diagram, then verify every label and relationship against the source notes.
- Keep the source references beside the final image so a reviewer can trace each visual claim.
Limits
The abstract does not give per-metric numbers, editing controls, or evidence that diagrams are publication-ready without review.
Related research
- MOSS-Image Technical Report This paper on arXiv Related paper on arXiv: Both papers generate structured visual documents with an explicit quality loop. PaperBanana plans, renders, and critiques academic figures, while the related system targets image generation with text-rich layouts and visual coherence.
Business ideas
1. paper figure service
Offer: A service that drafts publication figures from cited sources and a manuscript brief.
Buyer: Researchers
Technical link: Use PaperBanana reference retrieval, planning, rendering, and iteration.
Evidence
- competitor: Adobe launched Firefly image vector and video generation in one application (year: 2025) (source)
- adoption: 9.5% of EU enterprises used AI to generate pictures videos or sound/audio (year: 2025 region: EU denominator: enterprises with 10 or more employees or self-employed persons in covered NACE sections) (source)
Differentiation: Adobe Firefly is the named visual competitor; add paper-specific figure checks.
First test: Render 25 labeled figures from five approved manuscripts.
Success measure: At least 85% pass author review without a redraw.
Constraints: Pilot only. Use customer-owned or licensed data, approved runtimes, and human review. Market evidence is contextual proxy evidence rather than TAM.
2. reference-grounded style planner
Offer: A planning board that records figure references, layout, and style decisions before rendering.
Buyer: Researchers
Technical link: Use PaperBanana reference retrieval, planning, rendering, and iteration.
Evidence
- competitor: Runway offers enterprise API access to video image and audio models (source)
- adoption: 9.5% of EU enterprises used AI to generate pictures videos or sound/audio (year: 2025 region: EU denominator: enterprises with 10 or more employees or self-employed persons in covered NACE sections) (source)
Differentiation: Adobe Firefly is the named visual competitor; add paper-specific figure checks.
First test: Give two authors 20 generated figures with citation links.
Success measure: Authors can verify 95% of visual references in under two minutes each.
Constraints: Pilot only. Use customer-owned or licensed data, approved runtimes, and human review. Market evidence is contextual proxy evidence rather than TAM.
3. chart illustration checker
Offer: A checker that compares charts and labels against source data and flags defects.
Buyer: Researchers
Technical link: Use PaperBanana reference retrieval, planning, rendering, and iteration.
Evidence
- competitor: Meshy provides text to 3D image to 3D texturing animation and API workflows (source)
- adoption: 9.5% of EU enterprises used AI to generate pictures videos or sound/audio (year: 2025 region: EU denominator: enterprises with 10 or more employees or self-employed persons in covered NACE sections) (source)
Differentiation: Adobe Firefly is the named visual competitor; add paper-specific figure checks.
First test: Run 30 chart-heavy figures through the checker with seeded label errors.
Success measure: The checker catches 90% of seeded label and axis errors.
Constraints: Pilot only. Use customer-owned or licensed data, approved runtimes, and human review. Market evidence is contextual proxy evidence rather than TAM.
4. figure revision desk
Offer: An editor that stores figure versions, reviewer notes, and targeted redraw requests.
Buyer: Researchers
Technical link: Use PaperBanana reference retrieval, planning, rendering, and iteration.
Evidence
- competitor: Adobe launched Firefly image vector and video generation in one application (year: 2025) (source)
- adoption: 9.5% of EU enterprises used AI to generate pictures videos or sound/audio (year: 2025 region: EU denominator: enterprises with 10 or more employees or self-employed persons in covered NACE sections) (source)
Differentiation: Adobe Firefly is the named visual competitor; add paper-specific figure checks.
First test: Have three researchers revise ten figures using the revision desk.
Success measure: The final approved figure arrives within three revision cycles for 80% of cases.
Constraints: Pilot only. Use customer-owned or licensed data, approved runtimes, and human review. Market evidence is contextual proxy evidence rather than TAM.
5. publication asset audit
Offer: An audit packet linking each published figure to sources, prompts, and approval decisions.
Buyer: Researchers
Technical link: Use PaperBanana reference retrieval, planning, rendering, and iteration.
Evidence
- competitor: Runway offers enterprise API access to video image and audio models (source)
- adoption: 9.5% of EU enterprises used AI to generate pictures videos or sound/audio (year: 2025 region: EU denominator: enterprises with 10 or more employees or self-employed persons in covered NACE sections) (source)
Differentiation: Adobe Firefly is the named visual competitor; add paper-specific figure checks.
First test: Audit 50 submitted figures against the journal style checklist.
Success measure: The audit flags 90% of seeded publication defects before submission.
Constraints: Pilot only. Use customer-owned or licensed data, approved runtimes, and human review. Market evidence is contextual proxy evidence rather than TAM.