Upscayl vs Crisp: the same engine, doing a different job
Let's start with the part most comparison pages would bury. Crisp bundles upscayl-bin,
the same ncnn build of Real-ESRGAN that Upscayl is built around, redistributed under its
BSD-3-Clause and MIT licences and credited in Crisp's NOTICES file. On a single photo, at the same
model and the same scale, you should expect the same result. There is no secret sauce and claiming
one would be easy to disprove.
So the real question isn't which upscales better. It's what happens either side of the upscale.
Upscayl is an image tool, and a good one
Upscayl is free, open source, actively developed, and genuinely excellent at what it does. Point it at a folder of photos, pick a model, get bigger photos. If that is your problem, it solves it completely and costs nothing, and you should use it.
What it does not do is video. That is not a shortcoming, it's scope. A video is not a folder of photos, and treating it as one is where the difficulty starts.
What a video actually needs around the model
Run an image upscaler over extracted frames and you meet the problems in roughly this order.
Audio. Frames carry none. You have to demux it, keep it aligned through any frame-rate change, and mux it back without drift.
Frame rate. Phone and screen-recorded footage is often variable frame rate, so the gap between frames changes during the clip. Extract to stills and that timing information is gone, which is how audio ends up half a second out by the end of a long take.
Interlacing. Older footage stores each frame as two half-pictures captured moments apart. An upscaler reads the resulting comb pattern as fine detail and sharpens it into something much harder to remove. It has to be deinterlaced first, and nothing about a folder of PNGs tells you that it should be.
Colour. HDR sources need a proper tone map on the way in and out, or the result is flat and grey.
Order. Deinterlace, denoise, upscale, colour. Get that wrong and each step makes the next harder.
Scale. A ten minute clip is eighteen thousand frames. Somebody has to manage disk, memory, resumability and the encode at the end.
None of that is upscaling. All of it is what makes upscaling a video work.
The comparison that's actually useful
| Upscayl | Crisp | |
|---|---|---|
| Price | Free, open source | Free with a watermark; $19 once to remove it |
| Photos | Yes, its whole purpose | Yes, 9 formats including HEIC and AVIF |
| Video | No | Yes, with audio, timing and colour handled |
| Upscaling engine | Real-ESRGAN (ncnn) | The same Real-ESRGAN (ncnn) |
| Editing | No | Trim, crop, grade, stabilize, captions, timeline |
| Plain-English requests | No | Type what you want |
| Runs offline | Yes | Yes |
The honest recommendation
If you are upscaling photos, use Upscayl. It is free, it is excellent, and Crisp would be charging you for an engine you can already run yourself.
If you are upscaling video, the model was never the hard part, and a tool that only wraps the model leaves you to solve the rest by hand. That is the work Crisp is actually selling, and it is worth being clear that this, not image quality, is the difference.
Upscayl and Crisp, in detail
Does Crisp really use the same upscaler as Upscayl?
Yes. Crisp bundles upscayl-bin, an ncnn build of Real-ESRGAN, under its BSD-3-Clause and MIT licences, credited in Crisp's NOTICES file. On a single image at the same model and scale you should expect the same output.
Then why would I pay for Crisp?
For video, and for everything around the model: audio kept in sync, variable frame rate normalised, interlacing handled before it gets sharpened, HDR tone mapped, the ordering of restoration steps handled for you, and a timeline to actually cut with. For photos alone, Upscayl is free and does the job.
Can Upscayl do video if I extract the frames myself?
You can upscale the frames, yes. You then have to reassemble them with the original audio, preserve the timing of a variable-rate source, and handle interlacing and colour yourself. It's doable and it is a real afternoon of work per clip.
Is Crisp just a wrapper around Upscayl?
For the standard upscale lane it uses that engine, and says so. The rest of the app is its own: the timeline editor, the plain-English layer, stabilization, grading, captions, the montage builder, and the generative Max restore lane, which is a different model entirely.
Download Crisp for Mac Free to try, one-time $19 to remove the watermark. Runs entirely on your Mac.