Photo upscaling
2×, 4× and chained multi-pass scaling with GAN-sharpness or PSNR-fidelity models. JPEG artifacts, noise and blur repaired in one pass.
Version 2.0 — device-adaptive engine
AI upscaling,
tuned to your hardware.
Pro Edges profiles your GPU, VRAM and CPU at launch, then selects the optimal model from a curated library of 30+ engines — Real-ESRGAN, SwinIR, RGT, NAFNet and more. Photos, anime and video, up to 4× sharper. Processing never leaves your machine.
4.0× · UltraSharp V2
30+curated models
4–16×resolution gain
3GPU vendors supported
100%local processing
Output quality
Produced by the shipping engine.
A 150×100 degraded input restored to 600×400 with Real-ESRGAN x4plus — the same models and pipeline included in the application.
Input · 150×100
Output · 600×400 · 4×
Capabilities
A complete upscaling toolkit.
Video pipelines
Frame-accurate batch processing through FFmpeg. H.264, ProRes, VP9 or PNG-sequence output at full upscaled resolution, audio preserved.
Restoration modes
Dedicated denoise (NAFNet, SCUNet), deblur (MAXIM) and face restoration (GFPGAN, CodeFormer) stages, chainable in canonical order.
Any-GPU acceleration
CUDA for NVIDIA, DirectML for AMD and Intel, CoreML on Apple silicon, and tile sizes tuned to available VRAM everywhere else.
Fully local by design
No accounts, no telemetry, no cloud rendering. Models are fetched once from open repositories; inference runs entirely offline.
Batch processing
Drop entire folders into a managed queue. The scheduler keeps the GPU saturated and recovers cleanly from interruptions.
The smart engine
Pick an outcome.
The engine picks the model.
At launch, Pro Edges builds a hardware fingerprint: GPU vendor, exact VRAM (read from the driver registry, so cards above 4 GB report correctly), free memory, CPU cores and system RAM. Every recommendation is re-ranked against that profile — heavy transformer models only appear on machines that can run them efficiently.
- NVIDIA CUDA — unlocks FP16 execution and flagship SwinIR / RGT models
- AMD · Intel DirectML — CNN architectures selected for shader throughput
- Laptops, 2–4 GB VRAM — SPAN, ClearReality and AnimeVideo v3 stay responsive
- CPU-only systems — a lightweight stack that still outperforms bicubic clearly
CPUno usable GPU
Low< 4 GB VRAM
Medium4–8 GB
High8 GB+
Hardware scan — detected at launch
$ proedges --scan gpu_vendor NVIDIA gpu_name NVIDIA GeForce RTX 4060 vram_total 8.0 GB vram_free 7.6 GB backend DirectML cpu 12th Gen Intel i7-12700F · 20 threads ram 31.8 GB score 63 / 100 tier HIGH recommended UltraSharp V2 quality 8.5 alt_fast SPAN 4x speed 9.0 alt_max SwinIR 4x GAN quality 9.5
This exact scan powers the Recommended badges in the application — nothing is hardcoded per device.
Model library
Thirty-plus engines. Curated, not dumped.
Each model carries quality and speed ratings plus hardware-fit metadata. The headliners:
| Model | Best for | Scale | Quality | Speed | Min VRAM |
|---|---|---|---|---|---|
| SwinIR 4x GAN | Flagship photo detail | 4× | 9.5 | 4.5 | 6 GB |
| RGT x4 | Texture-preserving upscale | 4× | 9.0 | 3.0 | 4 GB |
| UltraSharp V2 | All-round crispness | 4× | 8.5 | 6.5 | 2 GB |
| Real-ESRGAN x4+ | Damaged, compressed photos | 4× | 8.0 | 6.0 | 2 GB |
| NAFNet SIDD | Heavy noise removal | 1× | 9.5 | 5.0 | 8 GB |
| MAXIM Deblur | Motion-blurred shots | 1× | 9.0 | 4.0 | 6 GB |
| CodeFormer | Face restoration | 1× | 8.5 | 4.0 | 4 GB |
| SPAN 4x | Fast quality upscale | 4× | 7.5 | 9.0 | 2 GB |
| AnimeVideo v3 | Anime frames, realtime rates | 4× | 7.5 | 9.0 | 1 GB |
Also included: GFPGAN, Swin2SR, SCUNet ×2, Real-CUGAN, NMKD-Siax, ArtClarity, NomosWebPhoto ATD, ClearReality and more. Models download on demand, typically 2–500 MB.
Workflow
Three steps. No timeline wrangling.
01
Import
Drag images or video files onto the window. Queue as many as you like.
02
Configure
Select modes — Upscale, Denoise, Deblur, Detail, Anime. Quality presets map onto your hardware tier automatically.
03
Export
Review with the comparison view, then export PNG, WebP, JPEG, MP4, ProRes or PNG sequences.
Install Pro Edges
Free. Windows 10/11 · Python 3.10+ · FFmpeg recommended for video.
- Run
ProEdges-Setup.batand follow the prompts — it installs to your user directory, no admin rights needed. - Launch from the Desktop shortcut. The engine detects your GPU on first start.
- Models matching your hardware download inside the app when first used.
The setup wizard is the official distribution channel and is verified on every release.
FAQ
Questions, answered.
Is my data ever uploaded?
No. All inference runs locally through ONNX Runtime or PyTorch on your own GPU or CPU. The application's only outbound connection fetches open-weight model files from their public repositories when you request a new model.
How does device-adaptive selection work?
On startup the engine profiles your machine: GPU vendor, total and free VRAM (read from the driver registry, so cards above 4 GB report correctly), backend availability (CUDA / DirectML / CoreML), CPU cores and system RAM. Each model carries quality, speed, architecture and VRAM metadata. Recommendations combine curated preference order with a hardware-fit penalty — for example, transformer models are deprioritized on CPU-only machines.
Are AMD and Intel GPUs supported?
Yes. DirectML works on any DirectX 12 GPU, covering modern AMD Radeon and Intel Arc/Iris boards. The engine steers those devices toward CNN-based models that run efficiently under DirectML.
What about low-end laptops?
Devices under 4 GB VRAM receive the light tier: SPAN 4x, ClearReality and AnimeVideo v3 produce strong results quickly with tiny downloads. Tiling adapts down to 128 px so memory limits are respected.
Can I chain multiple enhancements?
Yes — combine Denoise → Deblur → Detail → Face restore → Upscale in one pipeline. Stages run in canonical order, each with the best model your tier supports.
Does video upscaling preserve audio?
Yes. Audio is demuxed before frame processing and remuxed into the finished file.
Why is the source code not public?
The application is distributed exclusively through the signed installer channel on this page. The installer repository is public and auditable; the development repository remains closed.