October 6, 2026
Convert a 24 fps video to 25, 30 or 60 fps: three methods, including AI
Speed up by 4%, duplicate one frame in four, or create the missing frames by interpolation: what each method does to take a video from 24 to 25, 30 or 60 fps, what it damages, the ffmpeg commands, and what AI conversion costs, measured on our own shots.
Frank HoubreFounder of Imaginode
Three ways to change the frame rate
To go from 24 to 25, 30 or 60 frames per second, you can play the video faster, repeat some frames, or create new frames between the existing ones; only the third keeps both the duration and smooth motion.
A 24 frames per second video holds 24 snapshots for every second. A 30 fps video holds 30. Going from one to the other means finding six frames that do not exist. There are only three ways to do it, and every piece of software, from the simplest to the most expensive, uses one of them.
The first is to invent nothing and play the video faster: the 24 frames of one second are shown in slightly less than a second. The second keeps the speed and repeats existing frames to fill the gaps. The third looks at two neighbouring frames and computes the one that should sit between them: that is interpolation, which AI now does far better than the old algorithms.
The right choice depends on the target. From 24 to 25, the first method is a European television classic. From 24 to 30, the second is what most platforms do without telling you. And as soon as you want smooth motion at 30, 50 or 60, you need the third. Why you would aim for one frame rate or another is explained in 24, 25, 30 or 60 fps: which frame rate to choose.
From 24 to 25: speed up by 4%
Playing a 24 frames per second video at 25 shortens it by 4% and raises the sound by about 0.7 semitone; it is the historical conversion of films shown on European television, and it neither creates nor duplicates a single frame.
Between 24 and 25, the gap is so small that European broadcasters long chose not to recompute anything: the film is simply played 4.17% faster. Every original frame is shown once, motion stays perfectly even, and a 100 minute film lasts 96. This is often called PAL speed-up.
The price is in the sound: played 4.17% faster, it rises by about 0.7 semitone. A deep voice sounds a little higher, music shifts slightly in key. Current tools speed up the audio without changing its pitch. With ffmpeg, it fits on one line: ffmpeg -i input.mp4 -filter:v "setpts=PTS*24/25" -r 25 -filter:a "atempo=25/24" output.mp4. On a 10 second test clip, all 240 frames are kept and the file lasts 9.6 seconds, sound included, at the same pitch.
This method does not fit when duration matters: music cut to picture, a shot synced to a voice-over, an ad with a fixed length. It does not work above 25 either: going from 24 to 30 by speeding up would give a film 25% shorter and cartoon voices. In those cases, you keep the duration and make new frames.
From 24 to 30: repeat one frame in four
Without interpolation, 24 frames become 30 by repeating one frame in four: duration is kept, but motion advances in regular jolts, a stutter that is very visible on pans and tracking shots.
Thirty divided by twenty-four is 1.25: for every four original frames, you need five. The simplest answer is to show one of the four twice. That is what the command ffmpeg -i input.mp4 -vf fps=30 output.mp4 does, what most editing software does when you drop a 24 fps clip into a 30 fps timeline, and what a platform does when it receives 24 and displays 30.
The result keeps the duration and the sound, but motion is no longer even: three normal steps, one step in place, three normal steps. On a talking face, you see nothing. On a drone pan, a lateral tracking shot or scrolling text, the picture moves in small jolts, several times per second. American television lived with a variant of this flaw for decades, 3:2 pulldown, which spreads the repetition across half-frames.
The images below show the difference on the same shot. At the top, four consecutive frames of the source at 24 frames per second. In the middle, the five matching frames after duplication: two of them are strictly identical. At the bottom, the five frames after AI interpolation: each one is different, and the village moves at an even pace.
From 24 to 30 or 60: create the missing frames
AI interpolation estimates how each area of the picture moves between two consecutive frames, then computes the in-between frames: duration and sound do not change, and motion stays even at 25, 30, 50 or 60 frames per second.
An interpolation model receives two neighbouring frames and works out, for each small area, where it went: the roof moved down three pixels, the wave shifted left, the hand turned. From this it derives motion vectors, then paints an in-between frame where each area has travelled part of the way. Older algorithms did the same with square blocks and left torn edges; current neural networks handle outlines far better, as well as what appears behind a moving object.
To go from 24 to 30, the model places five frames where there were four, at moments that almost never fall on an original frame. For 60, it places two and a half per original frame. For 25, only one frame in twenty-five is truly invented, and the conversion is practically invisible. In every case, the video lasts exactly as long, and the audio track can be copied as is.
We measured it on two of our own shots generated at 24 frames per second, at 1280 × 720, on 6 October 2026. A drone flyover of 10.04 seconds, 241 frames, came out as 251 frames at 25 (10.04 seconds), 300 at 30 and 600 at 60 (10.00 seconds). A perfume ad of 8 seconds with its music came out as 240 frames at 30, sound intact and in sync. What can go wrong with interpolation, and when, is covered in AI frame interpolation.
Step by step in Imaginode
In the canvas, the Enhance node in Video mode changes the frame rate: Topaz at ×1 keeps the resolution and costs 6 credits per five seconds at 720p up to 30 frames per second, 12 at 50 or 60; the source duration and sound are kept.
Connect your shot, a Video node that just generated or an mp4 imported into a Media node, to an Enhance node in Video mode. Pick Topaz Proteus and the ×1 factor: resolution does not move, only the frame rate changes. Pick the frame rate, 24, 25, 30, 50 or 60, and run. The price is shown on the button, computed on the real length of the shot, one to sixty seconds per pass.
The rate follows the provider's: up to 30 frames per second, it is the price of an upscale, 6 credits per five seconds for a 720p output, 12 at 1080p. At 50 or 60, it doubles, because twice as many frames are computed. If your shot also needs more resolution, the ByteDance Upscaler does both at once: a 720p brought to 1080p and to 30 frames per second costs 5 credits per five seconds, 9 at 60.
The "Change a video's frame rate" template holds this workflow already wired, with a village flyover at 24 frames converted to 30 and to 60, for 18 credits for both passes. Replace the source with your shot and run it again. To upscale without touching the frame rate, the same logic applies in sharpen a video.
In editing software or with ffmpeg
DaVinci Resolve and Premiere Pro can interpolate with an optical flow mode when a clip's frame rate differs from the project, and ffmpeg offers the minterpolate filter; these local tools are free or already paid for, but slower and more prone to tearing than recent models.
In DaVinci Resolve, a clip whose frame rate differs from the project's is converted by default in Nearest mode, which means duplication. The clip's retime setting also offers Frame Blend, a dissolve between neighbouring frames, and Optical Flow, an interpolation; the more advanced Speed Warp mode is reserved for the Studio version, at 295 dollars. Premiere Pro, at 22.99 dollars a month, offers the same choice in the clip's time interpolation: Frame Sampling, Frame Blending or Optical Flow. In both cases, the computation happens on your machine, at render time.
With ffmpeg, the minterpolate filter does motion-compensated interpolation, without a neural network: ffmpeg -i input.mp4 -vf "minterpolate=fps=60:mi_mode=mci:mc_mode=aobmc:me_mode=bidir:vsbmc=1" output.mp4. It is free, but slow, the work runs on the processor, and fast objects leave torn edges or halos. The documentation itself warns that a scene change makes the motion vectors random.
These tools remain useful for a simple shot, a slow movement, a 24 to 25 conversion. For a shot with a lot of movement, a drone, a dance, a crowd, recent interpolation models give much cleaner outlines. The comparison of tools, free and paid, is in AI frame interpolation.
Three traps to avoid
Converting an edit that is already cut creates a ghost frame on every cut, aiming for 60 frames per second removes the cinema look, and confusing frame rate with slow motion means waiting for an effect that never comes.
First trap: converting the final video rather than each shot. On every hard cut, the model looks for motion between two unrelated frames and builds a frame where both shots overlap, visible for an instant at each cut. Convert shot by shot, then edit. If you only have the final file, some tools detect scene changes and skip the cuts; check frame by frame.
Second trap: putting everything at 60. A 60 frames per second video has the smooth look of a football match or a video game. That is perfect for sport, a screen tutorial, a product scroll. It is often poorly received on a fiction shot, where the slight motion blur of 24 is part of the aesthetic, hence the criticism aimed at TV motion smoothing.
Third trap: expecting slow motion. Going from 24 to 60 keeps the duration: the scene plays at the same speed, with more frames. Slow motion means playing those 60 frames at 24 per second, which makes the shot two and a half times longer. For real, spectacular slow motion, suspended drops and floating hair, it is better to generate it directly: see slow motion video with AI.
The summary
From 24 to 25: speed up by 4% if duration and sound may change, otherwise interpolate. From 24 to 30, 50 or 60: interpolate, shot by shot, before editing. Duplicating frames only works when motion is slow.
Duplication is free and instant, but it makes anything moving steadily stutter. Speed-up is perfect for a film at 25, at the cost of 4% of duration and a slight rise in sound. Interpolation is the only method that keeps duration, sound and even motion at the same time, at any frame rate.
In Imaginode, it costs 6 credits per five seconds at 720p up to 30 frames per second with Topaz at ×1, or 5 credits with the ByteDance Upscaler if you also go up to 1080p. Above 30, the price doubles. Sound is kept in both cases.
Almost every AI video comes out at 24 frames per second: if your edit, your channel or your screen expects something else, conversion is part of the job, like grading or upscaling. For the latter, see generate in 4K or upscale afterwards.