Research communication guide
How can I turn my research paper into a short explainer video with AI?
Updated by the Vidi team
5 free credits: two trial videos, 15 seconds each. No credit card required.
Making a presentation video by hand can take hours of slide design, recording and editing, even for a video of 2 to 10 minutes. AI tools now draft the script, narration and visuals from your paper. The work shifts from production to two other jobs: deciding what the video is for, and checking what the AI wrote.
This guide covers both. You will see how to choose a length and an audience, and how to give each scene one idea. You will also see which kinds of tools exist and who each one suits, and which mistakes to avoid before you post a video with your name on it.
How it works
How to do it with Vidi
Bring the paper or the abstract
For a full paper, open Research with Vidi. Upload the PDF (up to 50 MB and 300 pages), or add the paper by DOI, arXiv ID, OpenAlex or PubMed ID, a link or its title. For a quick explainer, paste your abstract or key findings into Vidi instead, and describe who the video is for.
Choose the format and check every claim
Choose a length from 30 seconds to 5 minutes, 16:9 or 9:16, and an AI voice or your own recording. Each claim is labelled supported, simplified, interpretation or analogy, and it must quote the page it comes from. Vidi checks that the quote is on that page and leaves unsupported claims out. You check that the meaning is right. For a pasted-abstract explainer, you review and edit the script scene by scene before it is animated.
Present and share it
Download the video with subtitles and a transcript, or share an unlisted link that you can switch off at any time. Use 16:9 for a talk or slides and 9:16 for social media. In Research with Vidi, a scene that rests on the paper shows the quote and its page number on screen.
Start with a brief, not the PDF
Before you upload anything, decide who the video is for and what it should do. Golpo's guide suggests a six-line brief: the paper title, a specific audience, the research question in plain language, the finding in plain language, why it matters, and what the viewer should do next. Ozor's guide also advises stating the audience, length and format in your prompt. Golpo's guide warns that a vague audience gives you a generic summary.
Then match the length to the job. Ozor suggests 60 to 90 seconds for social media and 3 to 5 minutes for YouTube and conferences. For comparison, author-made conference talks in the Paper2Video benchmark average about 6 minutes 15 seconds.
A one-minute video can carry one result, one figure or one caveat. A longer one can cover the question, method, result and limitation. If you need both, one NotebookLM guide recommends making a separate short video rather than cutting down the long one.
Structure it one idea per scene
Give each scene one idea, one picture and one spoken line. A simple order to try: the question, why it matters, what you did, what you found, the main limitation, and where to read more.
For short videos, the same NotebookLM guide suggests one audience, one question, one claim and one ending. It also advises you to protect one proof point and one caveat, and to end by pointing viewers to the paper.
Give the tool more than the abstract if you want evidence on screen. KPainter advises uploading figures, methods and limitations so the tool can show the evidence. Keep captions on, so the video can still be followed with the sound off.
There is a side benefit too. One author used NotebookLM's video of her paper to test whether the paper itself was clearly signposted and structured. If the AI misses your main point, readers might miss it as well.
Choose the kind of tool
There are three broad routes. The first is a narrated explainer with no presenter. Google NotebookLM Video Overviews makes a narrated slideshow from your uploaded sources and lets you set the audience and focus. Dedicated tools such as Golpo, Ozor, Scholarly and Vidi turn a paper or abstract into a narrated video, and SciSpace makes short, captioned video abstracts.
The second is a video with an AI presenter. Synthesia is one option, and that route is the better choice if you want a face on screen. The third is a video you design yourself. Canva gives the most control but needs more hands-on design work, and you may need to write the script first.
If you are technical, research pipelines exist too. PaperTalker generates slides, subtitles, a talking head and speech in the author's voice. It is heavy, though: its reported runs used eight NVIDIA RTX A6000 GPUs, and it was tested only on machine learning, computer vision and NLP papers.
For a low-cost first draft in English, NotebookLM is a sensible place to start. The speed and quality claims for dedicated tools come mostly from the vendors themselves.
Check accuracy and avoid common mistakes
Treat the output as a first draft. One NotebookLM user wrote that its video might oversimplify and miss the nuances. TechRadar warned that a minor mistake can be significant when analysing documents. Watch the whole video and check every number and every description of a figure. Look for your hedges too, such as 'may', 'in mice' or 'associated with', and make sure they are still there.
Mistakes to avoid: publishing without checking, losing the limitations when you shorten, and leaving the audience vague. Avoid forcing a fixed runtime too, which, as ngram puts it, can squeeze out the detail people came to understand. Do not expect one video to suit both social media and a conference. And do not assume a free tier includes every feature: NotebookLM's Cinematic mode is not on the free plan.
Before you upload an unpublished or embargoed paper to any tool, check your institution's and publisher's rules.
Ways to turn a paper into an explainer video
| Vidi (Research with Vidi) | NotebookLM Video Overviews | Dedicated paper-to-video tools | Design it yourself (Canva) | |
|---|---|---|---|---|
| What you give it | A PDF up to 50 MB and 300 pages, or a DOI, arXiv ID, OpenAlex or PubMed ID, link or title. Or paste an abstract into a Vidi explainer | PDFs, Google Docs, Google Slides, web pages and YouTube links | Varies: Golpo takes a PDF plus a short brief; Ozor takes the abstract, findings and a methods summary; researchpapervideos.com takes a PDF link | You may need to turn the paper into a script first |
| What you get | A narrated video with subtitles and a transcript; quotes and page numbers on screen; 16:9 or 9:16; AI voice or your own recording | A narrated slideshow mixing AI visuals with images, diagrams, quotes and data from your documents; a paid Cinematic version is fully animated | Examples: a whiteboard video (Golpo), captioned video abstracts (SciSpace), an MP4 with subtitles and music (YouMind) | The most manual design control, with more hands-on design work |
| Length | 30 seconds to 5 minutes | Short versions run about a minute; one tester got an 8-minute video from a 40-page paper | Golpo about 4 minutes; Scholarly about 2 minutes free, up to 45 minutes on paid plans | Up to you; AI clips work better for single scenes than long explanations |
| Checking claims | Each claim is labelled and tied to a quote and page; unsupported claims are left out; you check the meaning | Works from your sources, but one user said it may oversimplify and miss nuance | KPainter says it keeps evidence, interpretation and limitations distinct; no independent accuracy tests found | You write and check everything yourself |
| Cost | 9 credits per paper project including one video (1 credit = 1 US dollar); new accounts get 5 free credits | Free plan included 3 video overviews; Google AI Pro at $20 a month gave 20 a day (TechRadar, may be out of date) | Scholarly gives one free credit for about 2 minutes at 720p; others not covered here | Not covered in our sources |
| Narration languages | English or Vietnamese for paper videos | English only at launch, with more planned | Not covered in our sources | Not covered in our sources |
| Best for | Researchers who want every claim traced to a page, for a talk or social media | A low-cost English draft where you set the audience and focus | Speed with little editing skill, accepting that quality claims are the vendors' own | Full control over design, if you have time for hands-on work |
Limits worth knowing
- AI drafts can be wrong. A quote shows where a claim comes from, not that the reading of it is right, so check the meaning, numbers and caveats yourself before you share.
- Vidi narrates paper videos only in English or Vietnamese. A full paper video costs 9 credits, more than the 5 free credits a new account gets, and Free and Starter videos carry a small Vidi logo.
- Speed and quality claims for most paper-to-video tools, including figures quoted on this page, come from the vendors. We found no independent accuracy tests.
- We could not confirm journal rules on AI-made video abstracts, AI disclosure, reuse of published figures, or the confidentiality of uploading unpublished papers. Check with your publisher and institution.
Sources
- Paper2Video (arXiv abstract v1)
- Paper2Video (arXiv HTML)
- Paper2Video review (Moonlight)
- PaperTalker / Paper2Video (Emergent Mind)
- Research paper to video (Ozor)
- Research paper to explainer video guide (Golpo)
- NotebookLM short video overviews guide
- Research paper to video (KPainter)
- How I would have studied with NotebookLM video and audio overviews
- NotebookLM review (TechRadar)
- Explainer video maker (ngram)
- Google launches Video Overviews for NotebookLM (PPC Land)
- NotebookLM adds video summaries (Forbes)
- NotebookLM Video Overviews and Studio upgrades (Analytics Vidhya)
- NotebookLM short video overview (XDA)
- NotebookLM Cinematic Video Overview guide
- Text to video (Scholarly)
- Turn a document into an explainer video (EDUCBA)
- PDF to video research summary (SciSpace)
- Explainer video (YouMind)
- researchpapervideos.com on Hacker News
How long should a research explainer video be?
Match it to where it will play. Ozor suggests 60 to 90 seconds for social media and 3 to 5 minutes for YouTube or conferences. Author-made conference talks in the Paper2Video benchmark average about 6 minutes 15 seconds. A one-minute video should carry a single claim, with one proof point and one caveat.
Is there a free way to turn a paper into a video?
Yes, with limits. TechRadar reported that NotebookLM's free plan included 3 video overviews, though this may have changed, and its Cinematic mode is paid only. Scholarly offers one free credit for about 2 minutes. Vidi gives new accounts 5 free credits, but a full paper video costs 9.
Should I upload the whole paper or just the abstract?
Use the abstract for a quick, single-finding video. Use the full paper when you want methods, figures and limitations on screen. KPainter advises including figures, methods and limitations so the tool can show the evidence. Vidi accepts either: paste an abstract, or add the full paper in Research with Vidi.
Will the AI get my figures and numbers right?
Not reliably. One NotebookLM user said its video might oversimplify and miss nuance, and TechRadar noted that small mistakes matter when analysing documents. Watch the full draft and check every number, figure description and hedge against the paper before you post it.
Can the video use my own voice or face?
Some tools allow it. The research system PaperTalker adds a talking head and speech in the author's voice, and Synthesia offers an AI presenter. Research with Vidi lets you choose an AI voice or your own recording. Check any disclosure rules that apply to you before using a cloned voice or face.
Do journals accept AI-made video abstracts?
We could not confirm this. Our sources did not cover journal or publisher rules on video abstracts or AI use. Read your target journal's video abstract guidelines and its AI policy before you make the video, not after, and ask the editorial office if anything is unclear.