Constant quality, variable size
CRF stands for Constant Rate Factor, and the name describes a philosophy: instead of telling the encoder how many bits to spend ("give me an 8 Mbps file"), you tell it how much quality to keep ("keep everything looking this good"), and it spends however many bits that takes. A quiet interview scene needs few bits to look great; a confetti-filled action shot needs many. CRF lets the encoder budget accordingly, scene by scene, frame by frame.
The practical consequence is that file size becomes unpredictable but quality becomes consistent. Two ten-minute videos encoded at the same CRF can differ in size by a factor of five, yet both will look equally good relative to their sources. For everyday encoding — converting a file to play somewhere, shrinking a recording, publishing a clip — consistent quality is almost always what you actually want, which is why CRF is the default mode of the x264 encoder and of Mediamorphy's converter.
Reading the scale
For H.264 (x264), CRF runs from 0 to 51, and the direction trips people up at first: lower means better quality and bigger files. The useful range in practice is narrow:
| CRF | What you get | When to use it |
|---|---|---|
| 0 | Mathematically lossless, gigantic files | Almost never — this is not "highest quality worth having" |
| 17–18 | Visually lossless: differences invisible in normal viewing | Archiving a master copy, footage you will edit again |
| 22–23 | Excellent quality, sensible size — the sweet spot | Default for sharing, uploading, general use |
| 26–28 | Good quality, clearly smaller files | Messaging apps, email, limited storage |
| 30+ | Visible softness and blocking in complex scenes | Only when size matters far more than looks |
A remarkably reliable rule of thumb: every 6 points of CRF roughly halves or doubles the file size. Going from 23 to 29 cuts the file in half; going from 23 to 17 doubles it. This makes CRF easy to reason about: decide whether the current output is "too big" or "not sharp enough," nudge the number a few points, and you know approximately what will happen.
CRF vs bitrate: when each one wins
The alternative to CRF is bitrate targeting: you demand a specific data rate and the encoder hits it whether the footage needs it or not. That is the right tool when the size or rate itself is the constraint — live streaming within an upload budget, or fitting a file under a hard limit like an attachment cap. It is the wrong tool for quality, because a fixed rate over-spends on easy scenes and starves hard ones.
Streaming platforms re-encode everything you upload anyway, so pre-compressing to some exact bitrate rarely helps. Upload the best-quality file you can tolerate storing and let the platform do its own compression. For a deeper look at what bitrate actually measures — and how it interacts with resolution and frame rate — see our bitrate, resolution, and frame rate guide.
What CRF cannot do
CRF controls how faithfully the encoder preserves what it is given; it cannot restore what is already gone. Re-encoding a blurry, over-compressed source at CRF 17 produces a large file of the same blurry video. Quality flows only downhill through an encoding chain, so the best results always come from converting the original file rather than a copy that has already been through another encoder.
Repeated re-encoding compounds this. Each lossy pass discards a little more information, and errors accumulate — the video equivalent of photocopying a photocopy. Convert once from the source to your final format and settings, rather than converting the output of a previous conversion. This phenomenon, called generation loss, applies to audio too, as covered in our audio formats guide.
Choosing a value without overthinking it
A simple decision procedure covers nearly every case:
- Not sure? Use 23. It is the default for a reason — most people cannot distinguish it from the source on most footage.
- Keeping this file long-term, or editing it later? Use 18 and accept the size.
- File came out too big? Add 3–5 points and re-convert from the original. Consider dropping the resolution a step too — for large savings, resolution and CRF work best together, as our file-size guide explains.
- Output looks soft or blocky? Subtract 3–5 points, again starting from the original.
Testing tip: rather than converting a full one-hour video to check a setting, trim out a 30-second section with both calm and busy scenes, convert that, and inspect the result. Mediamorphy's trim control exists partly for exactly this — dialing in quality on a short excerpt before committing to the long encode.
One number, most of the control
Codecs are intimidating, and video engineering genuinely is deep — but the day-to-day levers are few. Where the file plays is decided by the container and codec; how big it is and how good it looks is decided overwhelmingly by resolution and CRF. Learn to nudge one number up when files are too heavy and down when they look too soft, and you know more about practical video encoding than most people ever need to.