Rice Upscaler
✨ Next-Gen AI Image Upscaler • Windows 10/11

Restore Ultra-Crisp Photo Details with
Rice Upscaler

Supercharge photos up to 8x resolution using DAT (DirectML) and Real-ESRGAN (Vulkan) neural networks. Hardware auto-cascade, exact target size fitting, fast batch processing, 100% offline & private.

⚡ Download Rice Upscaler (.exe) ⚙️ System Specs & Requirements

Release v1.0.0 • Windows 10/11 (64-bit) • 100% Local Offline Processing

Engineered for Peak Quality & Speed

Authentic capabilities directly mapped to the desktop application's real processing engines.

🧠

Intelligent Multi-Engine Cascade

Prioritizes Dual Aggregation Transformer (DAT/DAT-light) via DirectML for photorealistic detail; cascades to Real-ESRGAN via ncnn-Vulkan across NVIDIA, AMD, and Intel GPUs; and uses Lanczos as a rock-solid CPU fallback.

DAT DirectML Real-ESRGAN Vulkan Lanczos Auto
🔍

Scale 1x–8x & Print-Ready DPI

Flexible integer scaling slider from 1x (enhancement, compression, format conversion) up to 8x super-resolution. Selectable output DPI presets: Default (source DPI), 150 DPI, or 300 DPI print grade.

Scale 1x - 8x DPI Default / 150 / 300
🎯

Target Size Override (1–24 MB) & Formats

Specify exact output file sizes between 1 MB and 24 MB: binary search fitting and byte padding guarantee exact size compliance. Accepts JPG, PNG, WEBP, HEIC with fine-tuned JPG (1–100%) and PNG (0–9) controls.

Target 1–24 MB HEIC, WEBP, PNG, JPG
📁

Batch Queue & Drag-and-Drop

Drag-and-drop batches of images with Windows Long Path (>260 chars) and Unicode support. Multi-threaded background workers, live thumbnail previews, progress status, cancel anytime, and 10-minute timeout safety per file.

Drag & Drop Batch Queue Long Path Support
⚡

Cross-Vendor GPU Acceleration

Optimized natively for NVIDIA GeForce, AMD Radeon, and Intel Arc/UHD Graphics via modern Vulkan API and DirectX 12 DirectML. Adaptive VRAM tile sizing prevents memory exhaustion.

Vulkan 1.1+ DirectX 12 Adaptive Tiling
🔒

100% Offline & Private by Design

All neural inference executes strictly on your local PC hardware. Zero images sent to any server, zero telemetry tracking, no cloud dependencies, completely private.

100% Local No Cloud Zero Telemetry

System Requirements & VRAM Allocation

Accurate engine dispatching rules implemented directly in the application's runtime.

Operating System
Windows 10 / 11
64-bit (Version 1607+), Per-Monitor DPI scaling support.
Processor (CPU)
Intel / AMD 64-bit
Intel Core i3 / AMD Ryzen 3 or higher (AVX2 instructions recommended).
System RAM
4 GB - 16 GB+
4 GB minimum (requires >= 2048 MB free RAM for AI). 8–16 GB recommended.
Graphics & API
Vulkan 1.1+ / DirectX 12
NVIDIA GTX 900+, AMD RX 400+, Intel UHD 620+ / Arc. DirectML support.

Automatic Hardware Cascade: VRAM ➔ Active Engine

Detected VRAM Primary Engine Tile Size (Real-ESRGAN) Dispatching Rule & Capability
≥ 8 GB (8000 MB) DAT / DAT-light (ONNX DirectML) Tile 512 Full-speed DAT inference. High throughput with 512 tile for extreme resolution scaling up to 8x.
6 GB – 7.9 GB DAT / DAT-light (ONNX DirectML) Tile 256 Eligible for full DAT (>=6 GB) or DAT-light. Real-ESRGAN tile set to 256 for stable memory margin.
4 GB – 5.9 GB DAT-light or AI Real-ESRGAN Tile 256 Runs DAT-light (>=4 GB VRAM). Automatically cascades to Real-ESRGAN if image >10 MP to avoid OOM.
< 4 GB VRAM AI Real-ESRGAN (Vulkan) Tile 0 (Auto) Adaptive auto-tiling via ncnn-Vulkan. Highly stable on entry-level GPUs and budget laptops.
Scale = 1x or No GPU Lanczos Resampling (CPU) N/A (CPU) Scale 1x activates instant enhancement/DPI/size fitting. Provides 100% crash-proof fallback if GPU is unavailable.

Ready to Supercharge Your Images?

Download the standalone installer, pre-bundled with all neural models and GPU acceleration. No complex setups needed.

⬇️ Go to Download Page