AI Image Noise Remover
Remove grain, speckles, and digital noise from photos free in your browser using the Swin2SR Image Restoration Transformer — a neural network from Microsoft Research trained specifically on compressed and degraded images. It denoises by reconstructing clean structure, not just blurring. 100% private, no signup, no watermark. Works offline after first model load.
DENOISING STRENGTH
Drop a noisy photo here or click to upload
Supports JPG, PNG, WebP — max 10MB
Denoised Result
How Swin2SR AI Image Denoising Works
This AI Image Noise Remover uses Swin2SR — a Swin Transformer V2-based image restoration model from Microsoft Research — running via Transformers.js in your browser. The model was trained specifically on compressed and degraded images, making it ideal for photos with digital noise, JPEG artifacts, and film grain. Unlike a simple bilateral filter that only blurs, Swin2SR understands image structure and reconstructs genuinely clean output — preserving faces, textures, and edges rather than just smoothing them away. The model downloads once (~50MB) and is cached for offline use. Last updated: March 2026.
The denoising pipeline: your photo is resized to an optimal processing size (max 512px input for performance), fed through the Swin2SR neural network, and output with reconstructed detail. A light post-processing bilateral filter pass is then applied on top of the AI output at the strength level you selected (Light / Standard / Heavy). If the AI model fails on a slow connection or older browser, the tool automatically falls back to a Canvas-only bilateral filter.
When to Use an Image Noise Remover
Noise appears most commonly in photos taken in low light (indoor evening shots, night photography), at high ISO settings (ISO 800+), screenshots from old videos, and heavily compressed JPEG files with blocking artifacts. Noise removal is the first step before upscaling or printing any noisy photo.
- Low-light photography: Night shots, indoor concerts, restaurant photos — high ISO always produces noise.
- Old scanned photos: Scanner grain and film grain are both types of image noise that Swin2SR handles well.
- JPEG artifacts: Over-compressed JPEGs show blocky patterns that the neural network smooths out while reconstructing original detail.
- Pre-upscaling prep: Remove noise before using the AI Image Upscaler for cleaner enlarged results.
Denoising Strength Guide
The three strength levels control the post-processing bilateral filter applied after the Swin2SR AI output. Light (no bilateral pass) preserves maximum AI-reconstructed detail — best for photos with only mild digital grain. Standard (radius 1, threshold 15) is the recommended setting for most noisy photos. Heavy (radius 2, threshold 15) applies an additional smoothing pass — best for very grainy high-ISO shots, but may slightly soften very fine texture like fabric or hair.
Tips for Best Noise Removal Results
For best results, start with Standard strength and compare the before/after. If noise is still visible, re-upload and apply Heavy. For photos with important fine detail (hair portraits, fabric textures), use Light to avoid over-smoothing the AI output. After denoising, use the AI Photo Enhancer to restore some sharpness, as denoising can slightly soften edges. For very noisy old photos, apply noise removal first, then restore with the AI Old Photo Restorer for full reconstruction of lost detail.