✨ AI-Powered · 100% Offline · For Blender 5.0+

Upscale images 4x
without leaving Blender.

Three state-of-the-art AI models. CUDA-accelerated super-resolution. One-time purchase, no subscriptions, no cloud uploads. Your images stay on your machine — always.

3
AI Models
Upscale Factor
100%
Offline
0
Subscriptions
Requirements & Important Info

Read this before you buy.

Make sure your setup meets these requirements and limitations.

🧩
Software
Blender 5.0
💻
OS
Windows 10 / 11 (64-bit)
🎮
GPU
NVIDIA + CUDA
📊
VRAM
12 GB+ for 4K output
💾
Disk
~4 GB
🌐
Internet
Not required

Important — please read

  • Restart Blender after install. The AI engine and bundled Python libraries need a full restart to initialize.
  • Images must be RGB only. Alpha channels, normal maps, depth maps, roughness, metallic, and any non-color data are not supported.
  • Windows only · Blender 5.0+ only (no support for 4.2 or older).
  • NVIDIA GPU with CUDA is required. AMD, Intel and Apple Silicon are not supported.
  • Supported input: PNG, JPG, JPEG, BMP, TIFF, WebP · Output: PNG (lossless, 8-bit).
  • 16-bit PNG/TIFF images are auto-converted to 8-bit before upscaling.
  • If you run out of VRAM, drop to a lower precision or use the Medium (Swin2SR) model.
Watch It In Action

From low-res to stunning — in seconds.

See Neural Upscaler X4 running live inside Blender. No cloud. No waiting. Just results.

See the Difference

Before & After — drag to compare.

Real results from Neural Upscaler X4. Drag the slider to reveal the original vs the 4× upscaled version.

Before upscaling After upscaling Before After 4×

Sample 1 — DAT Model (Ultra)

720p upscaled to 4K

Before upscaling After upscaling Before After 4×

Sample 2 — DAT Model (Ultra)

720p upscaled to 5K

Pixel-Peeping · 5× Zoom

Look closer. See every detail recovered.

The exact same samples, zoomed 5× into the center. Drag to reveal the fine detail AI restores that simple scaling can't.

Before upscaling (5x zoom) After upscaling (5x zoom) Before · 5× After · 5×

Sample 1 — DAT Model (Ultra) · 5× Zoom

Pixel-level comparison of the center crop

Before upscaling (5x zoom) After upscaling (5x zoom) Before · 5× After · 5×

Sample 2 — DAT Model (Ultra) · 5× Zoom

Pixel-level comparison of the center crop

Features

Everything you need for professional upscaling.

Built from the ground up for Blender artists who demand quality, speed, and privacy.

🚀

GPU-Accelerated

PyTorch 2.7 + CUDA 12.8. Optimized for RTX 40 and RTX 50 series cards.

🔒

100% Offline

Your images never leave your machine. No cloud uploads, no data collection.

Background Processing

Keep working in Blender while the AI upscales your images in the background.

📁

Batch Processing

Drop a folder of images, get 4× upscaled results. Perfect for sequences and iterations.

🎯

Precision Control

Choose float32, bfloat16, or float16 to balance quality and VRAM usage.

💎

One-Time Purchase

No subscriptions. No recurring fees. No usage limits. Yours forever.

AI Models

Three models. Every quality need covered.

All three bundled inside. Pick the right tool for the job with a single click.

Medium
Swin2SR

Fast & balanced

  • Fastest processing
  • Ideal for large batches
  • Perfect for previews
  • Lower VRAM usage
High
HAT

Sharp & detailed

  • Superior detail recovery
  • Best all-around choice
  • Photorealistic results
  • Balanced speed & quality
Ultra
DAT

Maximum fidelity

  • Highest quality output
  • Finest detail preservation
  • Print-ready results
  • Final delivery grade
Precision Guide

The right precision for every resolution.

Save VRAM, go faster, keep quality — the built-in precision selector has you covered.

Input Resolution Output (4×) Recommended Precision
720p (1280×720)~5K (5120×2880)float32 — max quality
1080p (1920×1080)~8K (7680×4320)float32 or bfloat16
2K (2560×1440)~10K (10240×5760)bfloat16
4K (3840×2160)~16K (15360×8640)bfloat16 (float32 may OOM)
8K (7680×4320)~32K (30720×17280)float16 (workstation GPU)
VRAM Usage

Know what your GPU can handle.

Approximate values for the High (HAT) model. Medium uses ~30% less. Ultra is similar to HAT.

Input Resolution Output (4×) VRAM — float32 VRAM — bfloat16 / float16
720p (1280×720)~5K (5120×2880)~14 GB~12 GB
1080p (1920×1080)~8K (7680×4320)~20 GB~16 GB
2K (2560×1440)~10K (10240×5760)~28 GB*~22 GB
4K (3840×2160)~16K (15360×8640)~48 GB*~36 GB*
8K (7680×4320)~32K (30720×17280)Not feasible~80+ GB*

* Requires workstation-class GPU.

GPU Guide

What your card can do.

Your GPU VRAM Max Comfortable Input (bfloat16)
12 GBUp to 720p (1280×720) — minimum to load the model
16 GBUp to 1080p (1920×1080) comfortably
24 GBUp to 2K (2560×1440)
48 GBUp to 4K (3840×2160)
80 GB+8K possible (workstation GPU)
Use Cases

Built for every kind of artist.

🎨

3D Artists

Render at lower resolution and upscale for final delivery — cut render times without sacrificing quality.

🖼️

Concept Artists

Restore and enhance low-resolution references with studio-quality detail recovery.

🏛️

Visualization

Product shots, architectural renders, print-ready output with razor-sharp detail.

🎬

VFX & Cinematics

Produce crisp 4K and 8K output from modest render budgets.

Why Choose Us

Neural Upscaler X4 vs Online Services.

Neural Upscaler X4 Online Services
One-time purchase Monthly fee
Works offline Cloud only
Your images stay private Uploaded
Unlimited usage Quota limits
Integrated into Blender External
Batch processingLimited

Crystal-clear 4× upscales.
Right inside Blender.

Stop paying for cloud credits. Stop waiting for uploads. Stop compromising on privacy. Neural Upscaler X4 puts professional AI super-resolution directly in your Blender workflow — where it belongs.