Record everything
Nothing is missed, but a session fills the phone and the good run sits somewhere in hour two. Finding it means scrubbing, and the file still needs cutting before upload.
TL;DR. The personal best always lands on the attempt you stopped recording, or on the one you never started recording because the last forty died in the first minute. If you run on a handheld, a console hooked to a CRT, or anything without a capture card, you also have nowhere to press record in the first place. A rolling buffer on a phone flips that order: the camera keeps the last 45 seconds (free) or up to 12 minutes (Pro) at all times, and you save the run after the final split, from a watch or with an open palm. Resets cost nothing, because nothing is written unless you claim it.
Grinding a category is mostly resets. A runner going for a 7 minute PB might start 60 attempts in an evening and finish four of them. Record the whole session and you get three hours of footage to scrub, several gigabytes gone, and a file you have to trim before anyone will watch it. Record attempt by attempt and you are reaching for a button between every reset, which breaks the rhythm you are trying to build, and you will skip it on exactly the run that matters.
Nothing is missed, but a session fills the phone and the good run sits somewhere in hour two. Finding it means scrubbing, and the file still needs cutting before upload.
Clean files, but one missed press and the record is gone. Most runners stop doing it after twenty resets, then hit the PB on attempt twenty-one.
The camera records into a loop that overwrites itself. Reset as often as you like. When a run finishes and it is worth keeping, save, and the last N seconds become a normal MP4.
If you run on PC or stream a console through a capture card and OBS, you already have a replay buffer: OBS Studio ships one in its output settings, bound to a hotkey. Use it. A phone earns its place in the setups that never reach OBS.
The original Game Boy, Game Boy Advance and DS have no video output at all. A retail 3DS needs a modified board to capture, and a Switch in handheld mode sends nothing to the dock. For these, a camera pointed at the screen is how most runners film, and many handheld boards write their rules with that in mind.
A good capture chain for retro consoles (upscaler, splitter, card) costs more than the console did. Plenty of runners practise on a CRT for the zero input lag and only film when a run is worth submitting. A phone on a tripod in front of the tube covers that, with no extra hardware in the signal path.
Some boards ask for a handcam on categories where input tricks or macros are a concern, and viewers like seeing hard inputs anyway. A phone aimed at the controller runs its own buffer, so when the run is good you save the hands to go alongside the capture recording.
Practising on a handheld on the sofa, on a train, or at a friend's place with a borrowed console. No laptop, no account, no Wi-Fi needed. The buffer runs on the phone alone and saved clips stay on it.
The buffer has to hold three things: whatever your rules want shown before the timer starts (power-on, file select, the reset), the run itself, and the few seconds between the last split and your save. Set it one step above that total. A 9 minute run with a 30 second start sequence fits in 12 minutes with room to spare. A 12 minute run does not fit in a 12 minute buffer, because the start will already have been overwritten by the time you save.
| What you are running | Buffer | Fits |
|---|---|---|
| Individual level, single trick, short glitch category | 45 s (free) | Runs up to about 35 s plus a short setup |
| Longer IL, one world or stage | 90 s | Up to about 75 s |
| Short full-game categories, segment practice | 3 min | Up to about 2 min 30 s |
| Many retro any% and glitched categories | 6 min | Up to about 5 min 15 s |
| Mid-length full-game runs | 12 min | Up to about 10 to 11 min |
| Anything longer | Not the right tool for proof | Record continuously; use the buffer for highlights and practice |
Every step above 45 s is Pro. File sizes at your resolution and frame rate are in the storage and buffer calculator; a 12 minute buffer at 1080p60 needs a gigabyte or more of free space per saved run.
For a long category, the buffer still does useful work. Leave it on 45 or 90 seconds while you play and save the skip you finally landed, the boss you one-cycled, or the death you want to study later. Those short clips are what you actually rewatch, and they never sit inside a 2 hour VOD.
Verification is a person watching your video and deciding whether they trust it. Make that easy. Rules vary by game, so read the rules tab for your board first; these points cover what most camera-footage rejections come down to.
Frame so the full display sits inside the shot with a margin, including any on-screen timer, HUD, or item counter. A cropped corner is the first thing a moderator notices. Use the camera angle and distance planner to find how far back the phone needs to be for your screen size and lens.
Mount the phone level with the centre of the screen, not tilted up from the desk. For handhelds, an overhead arm pointing straight down works best. Turn off the lamp behind you: a reflection across a Game Boy screen hides exactly the pixels a moderator wants to see.
Most retro games run at close to 60 frames per second, so film at 60 fps to catch every visible change. On a CRT, expect some rolling bands; 60 fps keeps them milder than 30. Leave sound on. Saved clips keep the room audio, which carries the game music, your button clicks, and the reset jingle, and some boards use audio as part of the check.
If the board requires power-on from a cold boot, a file select, or a reset shown on camera, that footage has to be in the saved clip too. This is why the buffer table above leaves slack. A run whose first ten seconds were overwritten cannot be submitted, however clean the rest is.
The trim editor is fine for cutting dead air before the start sequence and after the final split. Never trim inside the run. A submission with a jump in it is the most common reason camera footage gets rejected, and it is hard to argue your way back once a moderator has seen a cut.
Both hands are on the controller until the last split, and your eyes stay on the game. The save has to come from somewhere else.
A Tile and a Complication on the watch trigger the save. Tap it once the timer stops. It takes under a second and you do not look away from the screen for the split comparison.
Put the controller down and hold an open hand toward the lens. It reads reliably at 2 to 5 metres in decent light. In a dark room lit only by a CRT, the watch tap is the safer trigger.
Fine when the phone is within reach, for example clamped next to a handheld. Try not to bump the mount: a camera shake right at the finish is the frame a moderator will look at hardest.
For a long evening, turn on stealth mode so the phone screen stays black while the buffer runs, and plug the phone into a charger or power bank. Without power, a phone recording continuously lasts roughly 60 to 90 minutes, depending on resolution. The battery estimator gives a figure for your phone.
Most of what a runner saves is not a record. It is the trick that failed for the fifth time, and you want to know why.
The Pro Analyze player steps one frame forward or back, plays down to 0.1×, and lets the phone's volume keys step frames. Put a failed attempt next to a good one in your head: where did the character stand on the frame you jumped, and where does the route guide say it should be?
Save the exact frame of a lineup and draw on it: a straight line at the pixel you need to reach, a label for the input. The result is one flattened image you can keep on the phone as a reference or send to someone in the game's Discord.
Point the phone at your hands and turn on Pro's delayed preview at 5 or 10 seconds. Do the input, glance over, and the screen is already showing what your thumbs did. Useful for a frame-tight buffer input or a mash you think you are doing evenly.
A phone filming a screen does not capture exactly one camera frame per game frame. Refresh rate, shutter timing and rolling shutter all shift things slightly. The footage is good for seeing what went wrong. For official timing and retiming, moderators still count frames on the source recording.
Pair the clips with numbers. The practice success rate tracker logs hits and misses per trick across sessions, so you can tell when a setup is consistent enough to put into full runs.
The saved file is a standard MP4 in the phone's storage. Nothing is uploaded, and there is no account to sign in to. When a run is good enough to submit:
That last step matters more than it looks. Twitch deletes past broadcasts after 7 days on most accounts, and a verification queue can take longer than that. Runners lose records every year because the only copy was a VOD that expired before a moderator got to it. A local MP4 does not expire. If you want the clip without the room noise for a highlight post, Pro can strip the audio track from a saved clip, but keep an unmuted copy for the submission.
Be clear about what this is. It is a camera app with a loop, built for sports coaches and adapted well to runs, not a speedrun tool.
The free tier covers a 45 second buffer, watch saving, slow-motion playback and trimming, which is enough for ILs and trick practice. Pro is a one-time unlock that adds the 90 s to 12 min buffers, open-palm saving, stealth mode, frame stepping, drawing on a frame and the delayed preview. There is no subscription.
Yes, and for handhelds with no video output it is often the only practical option. Mount the phone above or in front of the screen, keep the whole screen in frame, and record at 60 fps. Each leaderboard sets its own video rules, so read the rules tab for your game and category before relying on phone footage for a submission.
ReplayR keeps 45 seconds on the free tier and up to 12 minutes on Pro (90 s, 3 min, 6 min and 12 min steps). The buffer has to cover the run, whatever your rules require before the timer starts, and the few seconds it takes you to press save. In practice that means complete runs up to about 10 or 11 minutes. Longer categories need continuous recording for the proof video.
Tap the ReplayR Tile on a Wear OS watch after your last split, or put the controller down and show an open palm to the lens (Pro). The buffer keeps what happened before the trigger, so the run you just finished is what gets written to storage.
It depends on the game. Moderators write the rules per board, and many handheld boards accept camera footage because capture is hard on that hardware, while some categories require capture-card quality, a visible timer, audio, or a handcam. The file ReplayR saves is a normal MP4 with sound, so you can upload it to YouTube and link it in the submission like any other video.
Probably not for the main feed. OBS Studio has its own Replay Buffer that does the same job for anything going through your capture card. ReplayR is for the setups OBS never sees: handhelds, a console on a CRT with no capture chain, a second handcam angle, or practice away from the desk.
Roughly. The Pro Analyze player steps one frame at a time and plays down to 0.1x, which is enough to study a setup or see where a trick failed. A phone filming a screen does not line up one camera frame per game frame, so for official retiming, moderators still work from the source footage and its frame rate.