Upscale 1080p to 4K 2.5x faster: what changed in AI Video Enhancer 2026.7.11
Watch the short version
Windows · free trial · pay once, no subscription
AI upscaling needs an NVIDIA RTX 20-series or newer graphics card. Denoise, stabilisation and frame interpolation also run on AMD and Intel.
AI Video Enhancer Studio 2026.7.11 shipped on September 22, 2026. On our test PC, upscaling and 60 fps jobs finish in less than half the time, the upscaled video is bit-for-bit the same as before, and a 4K job no longer writes gigabytes of temporary frames to your disk.
Every number below comes from one PC: an NVIDIA RTX 3080 with a 24-thread CPU, old and new version run back to back on the same phone clips. The PC was doing other work at the same time, so a single timing can vary by about 20 %. Your times will differ with your hardware.
The speed-up at a glance
| Job | Before (2026.7.10) | After (2026.7.11) | What it means for you |
|---|---|---|---|
| Upscale a 15 s 1080p phone video to 4K (x2) | 69.6 s | 27.0 s | 2.5x faster. No temporary frame files (2.0 GB before). |
| Enhance (no upscale) a 15 s 4K 10-bit HEVC phone video with RTX VSR | 282 s | 132 s | 2.1x faster. 20 GB of temporary frames before, none now. CPU time from 1335 s to 451 s: a third of the processor work. |
| Enhance (no upscale) an 8 s 4K portrait phone video | 156 s | 59 s | 2.6x faster. 8.9 GB of temporary frames before, none now. |
| Convert a 15 s 1080p video from 30 to 60 fps (RIFE) | 205 s | 65 s | 3x faster, identical frames. |
| Convert an 8 s 4K portrait video from 30 to 60 fps (RIFE) | 337 s | 146 s | 2.3x faster. On PCs with less than 16 GB of RAM, 4K keeps the old speed. |
Before and after are the same program with the new code switched off and on. The upscaled video and the 60 fps frames came out byte-for-byte identical.
Why it is faster
Frames no longer go through your disk
Before, the app saved every frame as a PNG picture, let NVIDIA RTX Video Super Resolution upscale the folder, then read the pictures back to encode them. For a 15-second 4K clip that meant 20 GB of temporary files. Now the frames go from the decoder through the RTX upscaler into the encoder in memory. We compared the finished video streams of five test jobs, including 10-bit HEVC 4K and a portrait clip: identical, byte for byte.
A rebuilt RTX upscaling helper
The small program that talks to the RTX card now reads, upscales and writes frames at the same time instead of one after the other. In a 1080p upscale it went from 11.4 to 21.6 frames per second, and the 15-second job from 28.6 s to 22.7 s. At 4K the decoder is the slower part, so 4K times stayed about the same in this step. The output pixels are unchanged.
RIFE uses more of your processor
Frame interpolation to 60 fps uses RIFE, a neural network that draws the in-between frames. The app now picks RIFE's thread setting from your RAM. On the test PC the RIFE step of the 1080p job fell from 151 s to 31 s, with identical frames. For 4K on a PC with less than 16 GB of RAM the old setting stays, because the faster one needs about 2 GB more memory.
Fixes you will notice
Portrait phone videos upscale correctly
Upscale-only jobs on videos filmed upright on a phone came out as a striped, garbled picture while the app still said Done. The app now reads the video's rotation wherever it sizes frames. If you upscaled a portrait video with an earlier version, run it again.
Sharper stabilized video
Stabilize used to save a compressed in-between copy and then compress it a second time. When you only stabilize (no upscaling or interpolation), the correction is now applied in the final encode. With upscaling or interpolation the in-between copy is now lossless, and the app checks your free disk space first. Against a lossless reference, the VMAF quality score (100 = identical) rose from 97.0 to 98.6 on a 1080p clip and from 98.5 to 99.5 on a 4K portrait clip, with files 0.1 % to 0.5 % larger. Peak memory for 4K stabilize fell from 6.2 GB to 2.5 GB.
4K on a PC with 8 GB of RAM
On PCs with less than 11 GB of RAM, 1440p and 4K jobs now run inside a memory budget. In a test where the app was told it had 8 GB and was held to a 6 GB memory limit, all five 4K job types finished: RTX upscale, 60 fps, enhance plus 60 fps, stabilize plus enhance, and stabilize. The trade-off: 60 fps jobs run about 2.4x slower in this mode. Below 16 GB of RAM the app also processes 1440p and larger files one at a time.
No more false 'may be corrupt'
When the ffprobe tool was missing from an install, every video was reported as possibly corrupt. The app now repairs the missing tool on its own, analyzes any file that was never analyzed before the job starts, and says 'not analyzed' instead of 'corrupt' when a file still cannot be read.
MKV and WebM jobs run to the end
In the app (not the CLI), some MKV and WebM files could fail at the last batch of frames. Fixed.
How to get 2026.7.11
- Already installed: when you start the app, a banner tells you version 2026.7.11 is available. Download it and install it over your current version. Go to the download page
- New to the app: download the free trial and run your own clip through it. Go to the download page
- Windows 10 or 11, 64-bit. Upscaling and Enhance use NVIDIA RTX Video Super Resolution and need an NVIDIA RTX graphics card.
Questions
Does the faster version change my output?
The upscaled video and the 60 fps frames are byte-for-byte the same as with 2026.7.10 in every case we compared. Stabilized video does change: it is sharper, because it is compressed once instead of twice.
Will my PC be this fast?
These timings come from one PC with an RTX 3080 and a 24-thread CPU. The new version does less work for the same job (no temporary frame files, less CPU time), but we have only measured this one PC.
Do I need an NVIDIA graphics card?
For upscaling and Enhance, yes: they use NVIDIA RTX Video Super Resolution, which needs an NVIDIA RTX graphics card.
Time 2026.7.11 on your own clip
Take a phone video, upscale it to 4K and time it yourself.
Download Free TrialFree trial with watermark. No account or credit card required.