You have a screen recording of a task, maybe a Loom someone made last year, and people need it as a guide they can read: numbered instructions with a screenshot for each step. To convert a video into a step-by-step guide, you need three things from it: a picture each time the screen changes, one instruction per picture, and a person who checks both. This post shows how to do that with AtriumKit, how to turn a Loom video into an SOP, when recording the task again is the better choice, and the other ways to do it, by hand included.
We make AtriumKit, so weigh what we say about it accordingly. Facts about Loom and the other tools come from their own pages, checked on November 12, 2026.
The short version
- Get the video as a file: MP4, WebM or MOV, up to 10 minutes. Loom videos have to be downloaded first.
- In AtriumKit, click New demo, pick Guided demo and click Upload a video.
- Check the scenes it finds: one per screen change. Remove extras, add missed moments.
- Let AI draft the steps, using what’s said in the video if you like. Then read every step.
- Publish. You get an interactive demo and a step-by-step guide with a picture per step, which you can share as a link, paste as rich text or Markdown, or save as a PDF.
When a video beats recording clicks, and when it doesn’t
AtriumKit has two ways in. The Chrome extension records the task as you click through it: every click is a step, captured from the page itself. Upload a video works from a recording you already have. If you can still do the task in Chrome, recording it is usually better. A video wins in a few cases.
| From a video | Recording clicks with the extension | |
|---|---|---|
| You need | A video file of up to 10 minutes | Chrome, and the task in a browser tab |
| Desktop apps | Yes: a video is pixels, so any app on screen works | No, the extension records the browser only |
| Picture quality | A frame of a compressed video: text is softer, and a 720p recording gives 720p pictures | Captured from the page, with its HTML kept |
| Where a step points | Guessed from where the cursor was, or what changed. Check each one | The exact element you clicked |
| Private data | Frames aren’t scanned. Blur before you record, or in the editor | Auto-blur emails and names runs in your browser before upload |
| Fixing a renamed label later | Edit text paints new words over the picture, one screen at a time | Edit the real page text, or Find & replace… across every screen recorded as HTML |
| When the layout changes | Upload a new video | Recapture the screen, or record again |
Use a video when:
- The recording already exists. A subject expert walked through the process on a call or in a Loom, and asking them to do it again costs more than tidying the result.
- The person who knows the task won’t install anything. They can record their screen with what their computer already has.
- The task isn’t in a browser. A desktop app, a remote desktop session or a tool that only runs on someone else’s machine.
- The narration explains why. What’s said while each screen is up can feed the step text.
Record clicks instead when you can open the app in Chrome yourself and the guide will live for a while. Sharper pictures, exact click spots and text you can edit later save more time than the recording costs.
How to convert a video into a step-by-step guide
1. Prepare the recording
- Trim it to the task. AtriumKit takes MP4, WebM and MOV files up to 10 minutes and 1 GB. Cut the small talk at the start, and split anything longer into one video per task.
- Pause on each screen. Scenes need the screen to hold still. AtriumKit samples about 600 moments per video: every 0.2 seconds in a 2-minute recording, about once a second in a 10-minute one. In long recordings, hold each screen for a second or two before clicking.
- Leave out the camera bubble if you can. Scenes are found where the screen holds still, and a moving face in a corner may get in the way. If scenes go missing, use the controls in step 3.
- Don’t show what shouldn’t be in a guide. Video frames aren’t scanned for personal data, and AI reads the pictures.
Recording something new? On a Mac, Shift-Command-5 opens the screen recording toolbar and saves a .mov file (Apple). On Windows, Windows logo key + Shift + R starts a video snip in the Snipping Tool (Microsoft).
2. Upload it
- Click New demo in the sidebar, or press Alt+N (⌥N on a Mac).
- Pick Guided demo. Clickable demos get a guide too, but a guided demo’s intro and closing cards open and end the guide, and its tooltips become the steps.
- Click Upload a video and drop the file.
The video is read in your browser and isn’t uploaded. Only the screenshots of the scenes you keep are, plus the sound if you choose to use it in step 4.
3. Check the scenes
AtriumKit lists the moments where the screen settled on something new: a page, an open menu, a dialog. Each picture is taken just before the screen changed, so it shows what to click next, and a coral box marks where the step will point.
- Drag Fewer … More toward More if small changes, such as a menu opening, are missing.
- At most caps the steps: 15 by default, up to 30.
- Click a scene to see that moment in the video. Remove one you don’t need with its ×.
- To add a missed step, play or scrub to the moment and click Add this moment as a step.
4. Let AI draft the steps
The AI copy and voice settings sit below the scenes: Write the copy with AI, a Tone, a Goal (“Issue a partial refund”) and a Language. If someone talks in the video, check Use what’s said in the video for the copy: the sound is transcribed, and what was said during each scene helps AI write that step. The transcript isn’t kept.
The dialog shows “Uses about N AI credits” before you start. Copy costs 1 credit per 20 strings plus 1 per 10 pictures AI looks at, and the sound costs 1 credit per 5 minutes. A guided demo from a 4-minute narrated video with 12 scenes, without an AI voice, comes to about 5 credits. The Free plan has 25 AI credits a month.
Click Create and write with AI, or Create demo to write the steps yourself. The editor opens while AI writes.
5. Check every step
AI drafts. You review. It sees the pictures and what was said, but it doesn’t know your rules, which scene was a mistake, or why a step matters.
- Move the highlight. Select each tooltip step and drag the box onto the right button if the guess is off.
- Fix the words. Edit each step’s Title and Body. Use the labels exactly as the screen shows them.
- Blur what’s private with the Blur tool.
- Rewrite it as instructions. Open the Guide tab and click Write guide copy to turn tooltip copy into plain steps, like “Click New invoice”, at 1 AI credit per 20 strings (about 2 for 12 steps). Or type your own in Title in the guide and Text in the guide. Both change the guide only.
- Drop steps readers don’t need with Hide from guide. The demo still plays them.
6. Publish and export
Click Publish. The guide lives at the demo’s address with /g/ in place of /d/, with a numbered step and a picture for each scene, the spot highlighted. The Guide tab has the exports: Copy → Rich text or Markdown, Download .md, and Print / PDF. Copy and Download use your latest draft; Print / PDF uses the published guide.
See step-by-step guides for everything the guide page does, and create a demo from a screen recording for the full reference.
Loom video to SOP: download it first
Loom videos are a common source. Converting one works the same way, with two extra steps at the start.
Check that you can download it. Loom’s help, checked on November 12, 2026, says that downloads work on Business and Business + AI for admins and creators, on Education for admins, and on Enterprise for admins and members. On the free Starter plan you can’t download, and neither can Creator Lite users on any plan (Loom). Videos made with variables can’t be downloaded, only the organizer can download a meeting recording, and Enterprise admins can turn downloads off for everyone.
- In Loom, open the video from your Library, open the menu above it and choose Download. You get an MP4.
- If it’s longer than 10 minutes, trim it or split it by task.
- Upload it to AtriumKit as above. Most Looms are narrated, so check Use what’s said in the video for the copy.
- Make the guide an SOP. Put the purpose, scope, owner and “last reviewed” date in the intro card, which opens the guide. Put exceptions and who to escalate to in the closing card.
Can’t download? Record the task again: with the extension if it runs in Chrome, or with your computer’s screen recorder if it doesn’t.
Loom can also do this itself. Loom AI workflows “turn your video into a text document based on the content of your transcript”, with templates that include SOP and Step-by-step guide, and “Loom will automatically incorporate screenshots from your recording into the relevant steps” (Loom). Workflows are part of Loom’s AI Suite, listed on its Business + AI and Enterprise plans (Loom pricing). They don’t run on videos without speech, and trimmed or stitched videos get no screenshots. If your team already pays for it and the video is narrated, try it first. AtriumKit is the better fit when the video is silent, when you want an interactive demo next to the guide, or when you want each picture to mark the spot to click.
What to do with the guide
- Link it. On the demo’s Share tab, the Step-by-step guide card has Copy. The guide follows the demo’s access: public guides can be indexed by search engines, unlisted ones ask not to be.
- Put it in your help center. Embed the interactive demo at the top of the article (Share → Embed) and paste the written steps under it. The interactive help center playbook walks through it.
- Paste it into Notion, Confluence or Google Docs with Copy → Rich text, which brings the pictures along. The highlight is drawn into each picture, so it still makes sense outside AtriumKit.
- Save a PDF. Print / PDF opens the published guide in your browser’s print dialog, with each step kept on one page. Choose Save as PDF there.
- Feed it to an AI assistant. Add
.mdto the guide’s link for a Markdown version.
A pasted copy doesn’t change when you publish again, so paste it again after big changes. The hosted guide follows every publish.
Other ways to turn a video into step-by-step instructions
By hand
Play the video, pause at each screen change, take a screenshot (Shift-Command-4 on a Mac, Windows logo key + Shift + S on Windows), and write the step under it. It’s free and works for any video. It’s also slow: a 5-minute recording can hold 20 screens, and every picture needs its own crop and highlight.
With ffmpeg
If you’re comfortable in a terminal, ffmpeg can save a frame at each big change:
ffmpeg -i recording.mp4 -vf "select='gt(scene,0.3)'" -fps_mode vfr step-%03d.png
It saves the first frame after each change, so each picture shows the result of a click rather than the button to click, and the opening screen is left out. Lower 0.3 to catch smaller changes, such as menus. On older versions, use -vsync vfr in place of -fps_mode vfr. You still write every step yourself.
Video to SOP converters and other tools
- Loom AI workflows write an SOP or step-by-step guide from the transcript, with screenshots, on its AI plans (above).
- Clueso’s SOP generator takes a video in “MP4, MOV, or WebM” and says it will “automatically create a clean, step-by-step SOP with screenshots”, with export to PDF, rich text, Markdown or HTML (Clueso).
- Whale imports a video with Format as Guide on by default, turning it into a checklist where each step links to its moment in the video. Its help suggests videos “ideally under 5 minutes” and files under 100 MB (Whale).
- Trainn splits an uploaded video into clips “based on the narration”, then Convert into Guide makes each clip a step with a screenshot and AI text. Its own post warns that video frames “are sometimes caught mid-transition” (Trainn).
Loom’s workflows write from the transcript, and Trainn splits clips by the narration. AtriumKit splits steps by what changes on screen, so a silent recording works, and narration only helps with the words.
Record the clicks instead
For web apps, a click recorder skips the video: each click becomes a step with a screenshot. Our step-by-step guide with screenshots post compares the ways to do that, and has a template. For the browser extensions that record clicks, and the permissions each asks for, see Chrome extensions that create step-by-step guides.
FAQ
Can AI turn a video into a step-by-step guide on its own?
It can draft one. In AtriumKit, AI writes each step from its picture and from what was said during that scene. It can’t know what you meant to show, which scene was a mistake, or why a step matters, so read every step before you publish.
How do I create an SOP from a video or screen recording?
Upload the video, keep one scene per screen change, let AI draft the steps, then add what makes it an SOP: purpose, scope, owner and review date in the intro card, exceptions and escalation in the closing card. Paste it into your SOP library with Copy → Rich text, or keep the hosted guide as the source.
How long can the video be?
Up to 10 minutes and 1 GB, with up to 30 scenes. Split longer recordings by task. A guide of more than 30 steps is hard to follow anyway.
Does it work if nobody talks in the video?
Yes. Scenes come from the picture, and AI writes from the screenshots. Narration adds context, and costs 1 AI credit per 5 minutes of sound.
Can I update the guide when the product changes?
Partly. Edit text paints new words over a renamed label, one screen at a time. Screens made from a video can’t be recaptured, because they have no page address, so a changed layout means a new video or a new recording with the extension. Keeping help center screenshots up to date covers the options.
Is it free?
On the Free plan you can publish 3 demos at a time, each with its guide, and use 25 AI credits a month. More is on Pro and Team.
Try it with one recording
Pick a narrated recording of one task, under 10 minutes, that people keep asking about. Create a free account, click New demo → Upload a video, and publish the guide. No extension needed. If you can still do the task in Chrome, compare it with a click recording of the same task and keep the better one.