Scientific Slides

Scientific Slides is a skill in K-Dense's Claude Scientific Writer (MIT) for conference talks, seminars, thesis defenses, journal clubs, and grant pitches. Its stance is that research talks should be visual and story-driven, with minimal text on screen and citations drawn from the literature. Version 1.5 generates slides as images with an AI model and then assembles them.

Key Features

• PDF workflow (default): plan every slide (title, points, visual), generate each one with generate_slide_image.py, and combine them with slides_to_pdf.py slides/*.png -o presentation.pdf. The skill's docs call the model Nano Banana Pro; the bundled script's default slug on OpenRouter is google/gemini-3.1-flash-image.
• Consistency protocol: every prompt repeats a formatting goal (colors, fonts, layout), and each new slide is generated with --attach pointing at the previous slide so the style carries through the deck.
• Real data on results slides: the agent is told to list figures/ or results/ first and attach your actual charts, rather than inventing visuals for results.
• PowerPoint path: --visual-only produces images without slide text, which you then place with the PPTX skill so titles and bullets stay editable.
• Talk craft references: structure, timing and pacing, data visualization for slides, prompt writing, and common pitfalls live in the references/ folder, and validate_presentation.py plus pdf_to_images.py support a visual review pass.

Use Cases

• A 15-minute conference talk built from an existing paper and its figures.
• Thesis defense slides with a consistent visual style.
• Quick lab-meeting decks where visual polish matters more than editability.

Pricing and Access

The skill is free (MIT). Slide generation calls OpenRouter with your OPENROUTER_API_KEY and is billed per image by OpenRouter; citation gathering uses Research Lookup, which bills through Parallel.

Getting Started

1. Install the plugin in Claude Code: /plugin marketplace add https://github.com/K-Dense-AI/claude-scientific-writer, then /plugin install claude-scientific-writer.
2. Set OPENROUTER_API_KEY.
3. Ask for a talk plan, generate slides with python scripts/generate_slide_image.py "<slide prompt>" -o slides/01_title.png, then combine to PDF.

Limitation: in the PDF workflow each slide is a flat image, so text is not editable and small labels can come out wrong. Check every number on a generated slide against your data. Generation also sends your prompts and attached figures to OpenRouter.

Frequently Asked Questions

Can I edit the text afterwards?

Not in the image-per-slide PDF. Use the PowerPoint path if you need editable text.

Does it support LaTeX Beamer?

The skill's description lists PowerPoint and LaTeX Beamer as implementation options, with details in its reference files.

Alternatives

• Reveal.js Presentations: HTML slides with real, editable text and code highlighting.
• consulting-pptx-skill: consulting-style PowerPoint decks.
• LaTeX Posters Skill: the same research content as a printed poster.

Conclusion

Scientific Slides is best when you want a visually strong deck quickly and can live with image-based slides. If you must edit text later or avoid sending figures to a third-party API, choose an HTML or PPTX route. More presentation tools are in the skills hub.