An image that happens to move, and Confluence files it as one
Confluence's own docs file GIF under images, next to JPEG, PNG, and BMP, and pitch it as a way to add some fun. The structural reading matters more: a GIF enters the page as an image, and images render in place. No play control, no decision; someone once filed a request with Atlassian just to be able to pause one. The loop is already running when the reader's eye gets there.
Attach an MP4, MOV, or AVI instead and Confluence gives it a player: a play button, and a reader who has to commit. The current editor can't even be told to autoplay one. Right for a nine-minute walkthrough. Wrong for the drag handle that only works past a threshold, where it's a gate in front of a three-second answer.
The runbook pattern is boring and it works: text carries the sequence, one short loop sits beside the step that text can't carry. One loop per hard step. Not one per step.
100 MB is the ceiling. It is nowhere near the budget.
Confluence Cloud defaults to roughly 100 MB per attachment; a site admin can move it either way in General configuration, under Attachment settings, in bytes. Data Center defaults to the same figure, as of this writing.
Every new hire, every on-call engineer at 3 a.m., every auditor with a checklist opens the page and pulls every image on it. Six loops at 8 MB is close to 50 MB per visit. Keep each under about 2 MB and the page still feels like a page. The levers, in payoff order:
- Trim first. Start just ahead of the click and stop when the interface settles. Four seconds, not twelve.
- 10 to 12 fps. Menus don't move fast. A 60 fps capture is mostly duplicate frames of a panel holding still.
- 64 colors. Admin UI is solid color and type. Go toward 128 only when a screenshot inside the frame bands.
- Crop to the panel. A whole-monitor capture squeezed into a page column shrinks the label you're pointing at to a smudge. Crop tight, export near the width you'll set.
The projected size in the converter updates with each dial, so you land under 2 MB before export. Still too heavy? The under-2MB recipe pushes harder, including Extra Compression, mild grain for a much smaller file.
Name the file like you're going to replace it, because you will
The panel you looped in March gets redesigned in September, and a how-to GIF of the old layout is worse than none: it teaches confidence in the wrong place. Confluence has a mechanism for this that few people use deliberately. Upload a file with the same name to the same page and it becomes a new version of that attachment; older versions stay in the attachments panel, downloadable but not restorable, so the swap is forward-only.
Name the export for the step it shows, invite-member-step4.gif, not Screen Recording 2026-09-25 at 10.14.03.gif. Save the edit as a project file so the trim, crop, and caption are waiting for the re-record. Next quarter: new capture, same project, same filename, drag it onto the page.
Then set the width in pixels rather than eyeballing the handles, add alt text so a screen reader user knows what step four does, and keep the caption short; its size and alignment are fixed.
Four times a GIF is the wrong block on this page
GIF suits a short, silent, self-explanatory interaction. Confluence has better blocks for everything else:
- You need to talk over it. A GIF has no audio, and Atlassian owns Loom. Type /loom on the page and the recording lands as an embedded player, voice intact. A narrated architecture tour belongs there.
- The procedure runs long. Past fifteen seconds the file balloons and the repeat turns into a rewatch. Attach the MP4 instead; Confluence plays MP4, MOV, and AVI on the page, and this converter exports silent H.264 MP4 if you want it cut first.
- The page gets printed. A PDF export keeps one frame of a GIF at best; Atlassian's knowledge base even records a Data Center export crashing on certain animated GIFs until Java was upgraded. For a runbook that lives on paper, a still with an arrow wins.
- The reader has to copy something. A command, a config key, an endpoint. A GIF of text is a picture of text, and nobody can paste a picture. That's a code block's job.
Your security team vetted Confluence. They didn't vet a GIF site.
The footage in a runbook is the footage least suited to a random converter: the admin console, a staging URL, a customer's record open in the CRM. Upload that to a free web tool and it now sits on someone else's disk, kept for however long their terms say, at a vendor who never went through procurement.
This converter has nothing to put in your outbound logs. The page downloads once, and from then on the video is decoded, trimmed, and encoded by the same processor that's rendering this sentence. If a capture caught a customer email, a shape overlay covers it before export and is baked into every frame. The tutorial GIF guide covers pacing a step; the Jira page handles the bug-report cousin of this workflow.