To make a two minute research presentation video for a conference, write a script of 260 to 300 words in five parts: the problem in one sentence, the gap in what was known, your one idea, the single result that proves it, and what a viewer should do next (read the paper, come to the poster, try the code). Spend about a third of the time on the idea and a third on the result. Draw the method as a simple flow instead of showing your dense figures, put the title card on screen for three seconds at most, add captions, and check the venue's length, resolution and file size limits before you render. The video's job is to make a peer want the paper, not to replace it.

A conference video is an advert, not a short talk

The most common mistake is taking a fifteen minute talk and cutting it down. You end up with the same slides at four times the speed: motivation, related work, method, results, limitations, all rushed, none understood.

A two minute video has a different job. The people watching it are scrolling a programme, deciding which posters to visit or which sessions to attend. They already know your field roughly. They want to know, quickly, what you found and whether it matters to their own work. Everything that serves that decision stays. Everything else (related work, ablations, the second dataset, the hyperparameters) belongs in the paper, and the video should send them there.

This is also different from explaining your work to the public. A public video starts from an everyday question and avoids every technical term; that approach is covered in how to explain your research to the public in a video. A conference audience expects your terms. They just do not want all of them.

What venues actually ask for

Lengths vary a lot, and the limit is usually strict. A few real examples, so you can see the range:

Rules change between editions, so read this year's call for your venue, not last year's. The things to look for: maximum length, whether the title card counts, resolution, file size, file format, whether captions are required, and whether a voiceover is allowed. A few venues have asked for previews without narration, because they are played in a room.

If your venue's limit is 30 seconds, this post still applies: use the same five parts and give each one sentence.

The structure, by the second

For a two minute limit, aim for one minute fifty. Silence at the end is free; going over is not.

The numbers assume about 150 spoken words a minute. Most researchers read their own script slower than that, around 130, so time yourself before trusting the word count.

Two parts carry the video. The idea, because it is what a peer will remember and repeat. The result, because it is why they should believe the idea. If you are short of time, cut the gap scene down to a single sentence before you cut either of those.

Worked example: pothole detection from bus phones

Say your paper shows that cheap phones mounted on city buses can map potholes as well as a dedicated survey vehicle. Here is a full script, about 280 words, which runs just under two minutes at a normal pace. (The study and its numbers are invented for this example; use your own paper's exact figures.)

The problem. "Cities find potholes the slow way: a survey vehicle drives every road once or twice a year, and the rest of the time they wait for complaints. By the time a pothole is reported, it has often been damaging tyres for weeks."

The gap. "Phone accelerometers can feel a pothole, and earlier work used cars and cyclists to collect them. But private drivers choose their own routes, so coverage is patchy, and every phone is mounted differently."

Your idea. "Buses fix both problems. They drive the same routes every day, and every phone in a fleet can be mounted the same way, on the same bracket. We fitted forty buses with ordinary Android phones. Each phone records vertical acceleration and GPS. A spike above a threshold becomes a candidate. Because the same bus passes the same spot many times a week, we only keep candidates that repeat on at least three separate days. One-off bumps, like a bag dropped in the aisle, never repeat."

The result. "Over eight weeks, the bus phones found 412 potholes on the routes they covered. A city survey of the same roads found 431. Of our 412, 389 matched a surveyed pothole, so most of what we flagged was real, and we caught them, on average, eleven days earlier than the first public complaint."

What next. "The paper, code and the bracket design are linked below. Come to poster 27 to see the raw signal from a real pothole."

Three things this script does deliberately. The gap is stated as two concrete problems (patchy routes, inconsistent mounting), so the idea can answer both by name. The method has one clever step (the repeat on three days filter) and the script says why it works in a single sentence about a dropped bag. And the result has a comparison built in, because a number without a baseline tells a peer nothing.

Show the idea as a picture, not your figures

Your paper's figures were made to be read slowly, at full page width, with a caption. In a video they are on screen for ten seconds, often on a phone. Almost none survive that.

Redraw the method as a flow. For the bus example: a bus, a phone on a bracket, a spike in a line, a calendar with three ticks, a pin on a map. One box per sentence of narration, appearing as the sentence is spoken. If you need more than five boxes, you are explaining the paper, not the idea.

Pick one result and simplify it. Two bars (412 and 431) with the matched number marked is clearer than your confusion matrix. If your key result is a curve, redraw it with only the two lines that matter, and label them directly instead of using a legend.

Keep words on screen short. Three to five words per label. The narration carries the sentences, the drawing carries the structure. Viewers who watch with sound off (common in a busy conference app) should still follow from the labels and captions.

Drawn scenes suit this because they build up in step with the narration, so the viewer's eye is on the part you are talking about. The same reasoning applies at greater length in how to make a short video explaining a scientific paper's method, which gives the method a whole video rather than thirty five seconds.

Make it

1. 1

   Read this year's call for videos

   Write down the length limit, whether the title card counts, resolution, file size, format, and whether captions or a caption file are required. Set your target ten seconds under the limit.

2. 2

   Write the five-part script with a timer

   Problem, gap, idea, result, what next. Read it aloud at your normal pace and cut until it fits. Use the exact numbers from the paper, with their comparison.

3. 3

   Paste the script and keep your wording

   In openCanviz, paste the script, choose Keep my wording so every number is narrated exactly as written, set the length to two minutes, and pick whiteboard, which suits a method flow.

4. 4

   Check every box, label and number

   Pause on each scene. Drafted diagrams sometimes add a step you did not take or round a number. Fix labels in the editor and delete anything you did not do.

5. 5

   Record your own voice if you can

   Peers like hearing the author. Record in a quiet room and replace the generated voice. If your recording is noisy, the generated voice reading your exact script is the clearer option.

6. 6

   Export, caption and check the file

   Add a title card of three seconds and a closing card with the paper link and poster number. Check length, resolution and file size against the call, and attach a caption file if one is required.

Before you upload

• Length, measured on the exported file. Not on your timeline. Exports sometimes add a second.
• Captions checked by eye. Automatic captions misspell technical terms, model names and author names reliably.
• Co-author sign-off. Every author named on the title card should have seen the video, especially the result scene.
• Rights. Images, logos and music need permission. A conference upload with an unlicensed track can be muted.
• Free-tier watermark. If the venue or your lab expects a clean file, plan for a clean export before the deadline, not after.
• The link works logged out. If you host the video yourself, open the link in a private window.

Common questions

Should I show my face? Not needed. A two second shot of you at the end, or a photo on the title card, is enough for people to find you at the poster. The time is better spent on the result.

Can I reuse my talk slides? Use them as a source, not as the video. Copy the one sentence from each slide that matters into the script, then let the drawing replace the slide. Slides designed for a fifteen minute talk carry too much text for two minutes.

What if the venue limit is 30 seconds? Keep the five parts and give each one sentence: problem, gap, idea, result, where to find the paper. Drop the method flow to a single drawn scene. CHI's own guidance treats the preview as an elevator pitch, which is the right frame.

Is a narrated video allowed for a live competition like 3MT? No. 3MT is a live spoken talk with one static slide, and animation or video is not permitted. A two minute video is still useful practice for it, because writing the script forces the same choices. For other formats where you present in person, how to make a video presentation for a seminar in 30 minutes covers the live version.

How do I make a longer version later? Start from the same script and extend the idea and result scenes with the detail you cut: the second dataset, the main limitation, one ablation. The two minute video then works as the opening of a lab talk or a recorded presentation.