No more building paper presentations from scratch: turn one paper into diagrams, a PPT, or a long-form article
Hand over a paper link, PDF, or pasted text, choose diagrams / slides / a long-form article, and get back shareable finished images, PPTX/PDF, or an HTML page.
Turning your paper into diagrams or a PPT: what it can do for you
paper-craft-skills is a skill that turns academic papers into diagrams, slides, or in-depth long-form articles, for anyone who needs to prepare paper presentations, group meeting talks, public lectures, or science explainers. Give it an arxiv link, a PDF file, or directly pasted paper text, and after reading the paper it follows the approach that matches the output format you want.
The hard part usually isn't reading the paper, but presenting it. A paper contains a great deal, and no one decides for you which few points to cover: drawing a method diagram from scratch, assembling a slide deck, or writing an explainer article each takes hours, and it's easy to end up just restating the original text. This skill splits those three jobs into three paths, each with its own standards.
It first reads the whole paper, then confirms with you what you want. The diagram path proposes a suggested number of images and what each should cover, and only starts drawing after you confirm the language, scope, and style; the slide path first settles the page count, style, and whether to use the paper's real figures, then generates page by page; the long-form path has you choose one of three writing styles.
What you get is a finished product you can use directly. The diagram path outputs images plus a companion set of diagram notes; the slides are generated as individual 16:9 pages and finally merged into .pptx and .pdf, with each page keeping its own generation description, so you can redo just one page; the long-form article is an HTML page with built-in typesetting for equations and flowcharts, readable on both phone and computer.
Both style and language are selectable. Diagrams come in a warm hand-drawn-notebook style and a paper-grade framework-diagram style; slides have four presets—restrained academic, business research, warm journal, and glass texture; long-form articles come in three styles: story-driven, academic, and concise. The language for the images and pages can be set to Chinese or English.
Who it's for: graduate students preparing group meetings, defenses, and reading groups, developers who need to explain a paper to a non-specialist audience, and people creating course or science-explainer content. Who it's not for: situations where you need to fine-tune every text box word by word with others in PowerPoint, and material dominated by tables and financial data with almost no method diagrams. Also, the slide path requires a working image-generation capability in the current environment; when none is available, it stops and explains rather than passing off locally drawn graphics as the real thing.
How it differs from other paper-to-PPT tools
There are usually several ways to do the same thing: other paper-to-PPT or paper-explainer tools, general-purpose AI chat, and doing it yourself. Each covers part of the job—general chat can answer any detail in a paper on demand, and template tools turn out slides quickly. What this skill offers in exchange comes down to three things.
First, it confirms before it starts. Template tools grab the title and start applying a layout, and general chat hands you an answer straight away; this skill reads the whole paper first, proposes a suggested number of images and what each should cover, and waits for you to confirm the language, scope, and style before generating. The cost is a few extra confirmation steps; what you get in return is not ending up with a pile of images and pages you can't use.
Second, it checks the code. When writing an in-depth paper explainer, it finds the paper's corresponding open-source repository, reads the core source code, identifies at least two correspondences between the paper's method and the code, and marks the code locations in the article; if the paper has no public code, it notes the code status rather than glossing over it. General chat usually writes only from the paper's main text.
Third, slides are made page by page, so they're edited page by page. Each page keeps its own generation description, so "make page 5 a bit more academic" or "page 7 has too much text" affects only that page; template tools, by contrast, swap content into a single fixed layout, where changing one thing often means reflowing the entire deck. It also plans up front which pages will use the paper's real figures and experimental curves directly, instead of drawing everything from scratch.
When to choose it: you want a presentation or shareable material you can hand out as is, and you're willing to spend a few minutes confirming scope before you start. When not to choose it: you need a PPT that can be edited collaboratively down to every text box, your material is dominated by tables and financial data, or you just want to quickly ask about one detail in a paper—regular PPT tools and general-purpose AI chat fit those scenarios better.
Frequently asked questions about paper diagrams and PPTs
8 questions
Can I upload a PDF directly?
Yes. It accepts arxiv or DOI links, local PDF files, and directly pasted paper text; the slide path also accepts Markdown, technical articles, and existing outlines. PDFs are read in segments, and it only starts analysis after confirming it has the full content—it won't work from just the abstract.
Does it support Chinese?
Yes. Before generating, the diagram path lets you choose Chinese, English, or bilingual; for slides, you specify the language together with the page count and style; and the long-form path outputs a Simplified Chinese HTML page that opens directly in mobile and desktop browsers.
Can it generate just one image?
Yes, at least 1. After reading the paper, it first proposes a plan, typically a cover image, a method overview diagram, several core-mechanism diagrams, and one key-results figure, up to a maximum of 10; you can keep just the overview diagram, or expand to 8 to 10 images to explain each mechanism in more detail.
Does it cover the paper's background and related work?
The diagram path doesn't. It draws only three things: the overall method flow, the internal structure of each core mechanism, and the single most convincing set of experimental results. It does not draw related work, background, or multiple similar ablation experiments—those are left to the long-form path to handle in text.
Can the exported PPT be edited in PowerPoint?
Not exactly editable. Each page in the slide path is a generated 16:9 image, finally merged into .pptx and .pdf; the text is essentially inside the images, and it only performs format conversion and merging—it doesn't retouch the text on the images. To fine-tune text boxes word by word, a regular PPT tool is a better fit.
If I'm not happy with one page, do I have to redo the whole deck?
No. Each slide page keeps the generation description from that run, so requests like "make page 5 a bit more academic" or "page 7 has too much text" change and regenerate only that page; the other pages are unaffected, and images already generated won't be overwritten.
Does it really read the whole paper, or just the abstract?
The whole paper. An arxiv link reads both the abstract page and the full-text page, and PDFs are read in segments until the full content is covered; when writing a long-form article, it also looks for the paper's open-source code, reads the core files, and identifies at least two correspondences between the method and the code—and it also notes it if the paper has no public code.
What kinds of papers is it suited to?
Papers centered on methodological innovation, with architecture diagrams or citable figures, are the best fit, and it's useful for group meetings, defenses, courses, and science explainers. If your material is dominated by tables, financial, or compliance data, or you just want to quickly look up one detail in a paper, it isn't the first choice.