Devlog GIF Capture Tools and How to Shrink the Files
The best clip we ever recorded was six seconds of a boss slamming into a wall and dropping its shield, and almost nobody saw it properly.

The best clip we ever recorded was six seconds of a boss slamming into a wall and dropping its shield, and almost nobody saw it properly. I exported it straight out of the recorder as a GIF, got a 24 MB file, and posted it anyway because it looked fine on my monitor. In the devlog thread it took eleven seconds to appear on a phone, and three people replied asking what was happening in the clip. The animation was good. The file was the problem.
Capture is the part of devlog work nobody warns you about. You spend a week on a mechanic, then lose the one chance to show it because the clip is too heavy to load, too short to read, or the colors turn into a swamp. Here are the tools we settled on after about thirty of these, the actual file sizes we measured on the same clip, and the habit I would change first if I were starting over.
Why six seconds turned into 24 MB
GIF is a format from 1987 and it shows. Every frame carries its own palette of at most 256 colors, and compression works mostly by storing what changed between frames. That second part is the trap for game footage. A dialogue box opening against a still background compresses beautifully. A screen shaking while a particle burst covers the whole frame means almost every pixel changed, so almost every pixel gets stored again.
Three dials control the size and they multiply rather than add. Resolution, frame rate, and duration. Our boss clip went out at 1280 by 720, 30 frames per second, 6 seconds. That is 180 frames of mostly fresh pixels. Drop it to 960 by 540 and 15 frames per second and you are asking the encoder to store roughly a quarter of the data before any other change.
The color limit causes the other complaint, which is banding. Pixel art survives the 256 color cap easily because it rarely uses more than forty. Anything with a gradient sky, a bloom effect or soft lighting does not, and you get ugly stepped bands unless the encoder picks a palette from the whole clip rather than frame by frame.
Most of your clips should not be GIFs at all
The standard devlog advice is to make a GIF because GIFs autoplay and loop and video does not. That was true around 2016. It mostly is not now, and following it costs you quality for no reason.
Our same six second clip, encoded as H.264 at 960 by 540, is 480 KB and looks better than the 3 MB GIF. That is not a small gap. Every social platform we post to accepts short video, loops it or near enough, and compresses it far more gracefully than it compresses a heavy GIF. A viewer on mobile data gets the video in a blink and the GIF after a wait.
So the rule we use now: video by default, GIF only where the slot genuinely requires an image file. Those slots are real and there are more of them than you expect. Store page screenshot galleries usually take images only. A README on a code host renders a GIF inline and will not autoplay a video. Some forum software and some older Discord embeds behave the same way. Chat apps where people hover and scroll past quickly also favor GIF, because there is no play button to press.
What I would not do again is make a GIF for a platform that supports video and then wonder why engagement on that post was flat.
What each tool produced on the same clip
| Tool | Platform | What it is good at | Our 6 second clip |
|---|---|---|---|
| OBS Studio | Windows, macOS, Linux | Clean MP4 at a locked canvas size | 480 KB as MP4 |
| ShareX | Windows | Region capture bound to a hotkey | 7.1 MB GIF at defaults |
| ScreenToGif | Windows | Editing and deleting frames after capture | 4.6 MB GIF |
| LICEcap | Windows, macOS | Tiny download, nothing to learn | 9.8 MB GIF |
| Kap | macOS | Good defaults, trim handles built in | 5.2 MB GIF |
| Peek | Linux | One window, one button | 6.4 MB GIF |
| ffmpeg two pass | Everywhere | Best size for the quality, scriptable | 3.0 MB GIF |
| gifski | Windows, macOS, Linux | Smoothest gradients of anything we tried | 2.8 MB GIF |
Those numbers come from one clip on one project, so treat the ranking as a hint rather than a benchmark. The gap between a default export and a two pass palette export held up across everything we tested, though, and it was never smaller than about half.
One practical habit: we record with OBS and convert afterwards, rather than recording directly to GIF. Recording to GIF locks in the resolution and frame rate at capture time, and you will want to change both at least once.
The two commands we actually run
Record an MP4 first. In OBS, set the base canvas to 1920 by 1080, the output scale to 960 by 540, and the frame rate to 30. Then trim the clip with ffmpeg -ss 00:00:12 -t 6 -i raw.mkv -c copy clip.mp4, which cuts without re encoding and takes under a second.
Then build a palette from the whole clip and apply it. First ffmpeg -i clip.mp4 -vf "fps=15,scale=960:-1:flags=lanczos,palettegen=stats_mode=diff" palette.png. Then ffmpeg -i clip.mp4 -i palette.png -lavfi "fps=15,scale=960:-1:flags=lanczos[x];[x][1:v]paletteuse=dither=bayer:bayer_scale=3" out.gif.
The reason this beats a one pass export is the palette. One pass picks colors frame by frame and the palette shifts as the clip plays, which is where the shimmering comes from. The diff stats mode weights the palette toward the parts of the frame that move, which is almost always the part you wanted to show.
If it is still too big, cut the frame rate before you cut the resolution. Twelve frames per second reads fine for most game footage and saves more than scaling down another step, and scaling down is what makes a clip unreadable on a phone. Running the result through gifsicle with the optimize level set to 3 and the lossy value around 60 usually takes another 20 to 30 percent off. On pixel art, skip the dithering entirely by setting the dither option to none, since dithering adds noise that the format then has to store.
Capture the right six seconds
A tight clip beats a clever one. The pattern that works for us is one action, one reaction, and a loop point that lands where the scene looks the same as it did at the start. If the first and last frames do not roughly match, the loop reads as a stutter and people watch it twice and leave.
Hide the mouse cursor unless the mechanic is a mouse mechanic. Turn off the debug overlay, the frame counter and anything showing a file path, because somebody will read it and ask about it instead of the game. Give the action about half a second of lead in so the viewer's eye lands before the thing happens.
Keep the raw recordings. We lost a month of good footage once by clearing a scratch folder, which is the same lesson as backing up a game project before a crash costs you a week, just cheaper. Sprite work gets reused the same way, and the frame timing habits carry over from our Aseprite walk cycle workflow. Once the clip is cut, it goes up with the build, which for us means one butler command rather than the upload form.
FAQ
What size should a devlog GIF be?
Aim under 3 MB and treat 5 MB as a hard ceiling. Below 3 MB it loads on mobile data before anyone scrolls past, which is the only threshold that matters.
Why does my GIF look washed out compared to the game?
Almost always the 256 color palette, picked per frame. Build the palette from the whole clip with the two pass method and the washout mostly goes away. If the game uses heavy bloom, consider reducing it just for the capture.
Can I record a Scratch project the same way?
Yes. Run the project full screen in the browser, capture the stage region rather than the whole window, and lock the capture to 480 by 360 so the stage pixels map one to one instead of being resampled.
Is there a reason to still use LICEcap?
Only convenience. It is a 500 KB download that does one job with no setup, which is worth something at midnight. The files it makes are two to three times larger than they need to be.
More in Tools
Tools
Running Godot 4 on a Low Spec School Laptop Without Lag
Nine of the fourteen laptops in the Tuesday class I teach are the same kind of machine: a four or five year old 14 inch model,...
Tools
Using butler to Upload itch.io Builds in One Command
We shipped a jam game last spring and then spent two days patching it.
Tools
How to Use a Kanban Board for Game Development
Six months into our second game, we had a 300 line to do list in a text file, three people working from different versions of...