How fast is Basecut? What we measured for 4.1.1
Basecut 4.1.1 is faster in four places we measured on a Mac: the color step of analysis, the scopes in the Color tab, editor exports with effects, and the still frames the AI looks at on HEVC clips. The output is identical to before. These changes do not speed up the AI model, which is the longest step of an analysis, so a full wedding is still an overnight job.
Basecut is our product. Measured between 2026-10-20 and 2026-10-24.
The numbers
Each row is one task, timed before and after the change on the same 4K footage. Read across a row, not down a column.
| What | Before | After | Test |
|---|---|---|---|
| Color step of analysis, 4K clips | 72 to 121 seconds | 29 to 45 seconds | Three 4K DJI test clips, median of 3 runs each. |
| Color tab scopes (waveform, parade, vectorscope), one 4K frame | 1.87 seconds | 0.07 seconds | Peak memory during the refresh fell from 1.2 GB to 0.6 GB. |
| Editor export, 4-second 4K clip with opacity, soft crop and a title | 51.7 seconds | 19.1 seconds | Median of 3. The same clip with no effects: 25.7 seconds before, 17.6 seconds after. |
| Editor preview with effects and a title | 28 frames per second | 109 frames per second | 960 pixel preview, render plus JPEG encode. |
| Still frames for the AI, HEVC clips on a Mac | 5.3 to 5.7 seconds per clip section | 0.9 to 1.1 seconds | Eight stills from each of three 4K 10-bit HEVC clips. Processor time fell from about 19 to 23 seconds to about 1 to 3 seconds. |
All timings come from development builds of 4.1.1 on an Apple Silicon Mac that was also running other work, so the absolute seconds are rough. When the still-frame test was repeated on a very busy Mac, the gain was between 2 and 4 times instead of about 5. Compare before with after inside a row.
Same output, less waiting
Each speedup replaces a slow stretch of Basecut's own number crunching with compiled Rust code. 4.1.1 is the first release that ships these compiled modules. The old code stays in the app as a fallback, and Basecut uses it if a module is missing or fails its built-in self-check.
We tested that the new code gives exactly the same result as the old. The color statistics and scopes match bit for bit. The editor's compositing matches pixel for pixel in all 12 export setups we compared. The HEVC still frames match byte for byte. ProRes clips did not match exactly on the hardware decoder, with up to 4 values out of 255 different, so they stay on the processor.
What did not get faster
- The AI model. In our profile of five 4K DJI clips, Basecut's own code was about 12 seconds of a 1,126 second analysis run. The AI model, video decoding and encoding, and numerical libraries make up the rest, and these changes do not touch them.
- A plain 1080p export of one clip. It took 10.1 seconds before and 10.5 seconds after, because video encoding sets the pace there, not compositing.
- H.264 stills. The hardware decoder gave the same pictures but was not faster on the clock, so H.264 clips stay on the processor along with ProRes. Only HEVC clips on a Mac use the hardware decoder.
- Windows. We have not run or timed 4.1.1 on Windows yet, so there are no Windows numbers on this page.
How we measured
Every number is the median of three runs of one task, with the old and new code run on the same files. The footage is 4K HEVC from a DJI Osmo. The Mac was busy with other work during most runs, so we compare within a row and say so when the result moved with load. Packaged, installed builds have not been timed yet. The rough view in 4.1.1 has its own page.
More guides
Questions people ask
How much faster is Basecut 4.1.1?
On a Mac, the color step of analysis on 4K clips dropped from a range of 72 to 121 seconds to a range of 29 to 45 seconds, the Color tab scopes from 1.87 seconds to 0.07 seconds, and a 4-second 4K editor export with effects from 51.7 seconds to 19.1 seconds. The still frames the AI looks at on HEVC clips arrive about 5 times faster (2 to 4 times when the Mac was very busy). The output is identical to before. These changes do not speed up the AI model, which is the longest step of an analysis, so a full wedding is still an overnight job. These are measurements on development builds of 4.1.1 on a busy Apple Silicon Mac, so treat them as rough. Windows has not been measured yet.
Do these speedups shorten the overnight cull?
Not by much. In our profile of five 4K DJI clips, Basecut's own code was about 12 seconds of a 1,126 second analysis run, around 1 percent. The rest was the AI model, video decoding and encoding, and numerical libraries that were already compiled, and these changes do not touch them. The speedups target the part that was slow, which is color math and compositing, so a full wedding is still an overnight job. How to cull wedding footage with AI has the run times we have measured.
Do the faster paths change my pictures or my color?
No. We compared each new path with the old one on the same inputs. The color statistics and scopes match bit for bit, the editor's compositing matches pixel for pixel in all 12 export setups we compared, and the HEVC still frames match byte for byte. Where a hardware path did not match exactly (ProRes), Basecut keeps using the old path.
Do I have to turn anything on?
No. The faster paths are on by default. If a compiled module is missing or fails its built-in self-check, Basecut falls back to the previous code, so the result is the same and only the speed differs.
Is Basecut 4.1.1 faster on Windows?
We do not know yet. The new code is built for Windows too, but we have not run or timed it there. The HEVC still speedup uses the Mac's hardware video decoder, so it applies to Macs only.
Time it on your own footage
The only fair test is your own card dump and your own computer. Basecut has a 14-day free trial with 5 free exports and no card.
Watch the demo