Photo Restorer
Restore old and degraded photos with AI — remove noise, blur and artifacts. Free, unlimited.
画像をここにドラッグ＆ドロップするか、クリックして選択
Supports PNG / JPG / JPEG / WEBP, up to 40MB
- 1枚の画像は通常5〜30秒かかります
- 結果は24時間保持され、その後自動削除されます
- タスクは自動保存されます：ページを閉じても24時間以内に戻って確認できます（最後のタスクのみ保持。シークレットモードやブラウザキャッシュ削除は避けてください）
処理完了後、結果がこのアドレスに送信されます（1回のみ）
Paint over areas to remove
Paint over the people, objects, watermarks or blemishes to remove — AI will fill the background. Cover the area fully for best results.
アップロード後、修復された画像がここに表示されます
How Old Photo Restoration Works
Old photo restoration starts with a generative adversarial network (GAN), a deep learning approach built from two competing neural networks. The generator learns from millions of clear photographs, building an internal model of what natural, high-quality detail looks like — skin texture, fabric weave, foliage, and fine edges. Given a damaged or low-quality input, it reconstructs the missing information with plausible, realistic detail. The discriminator acts as a critic: it examines each restored image and tries to decide whether it is a real photograph or an AI-generated result. Through this adversarial training loop, the generator is pushed to produce images that are not only sharp but also natural-looking, matching the statistics of genuine photos. Before reconstruction, the pipeline also applies automatic denoising and deblurring that remove film grain, dust, scratches, and motion blur, and it cleans up JPEG compression artifacts. The final output is a cleaner, sharper picture that preserves faces and important details without looking over-processed.
Who It's For
You can restore old photos free in many everyday situations. Digitize and refresh family albums: faded, creased pictures of parents and grandparents can be cleaned up before printing, framing, or sharing with relatives. Repair blurry scans: photos and documents scanned years ago with low-resolution equipment often come out soft, noisy, or slightly smeared, and a quick pass brings back sharpness and readability. Clean up compressed images: pictures shrunk for messaging apps or social media lose clarity, so restoring removes blocky compression artifacts and recovers crisp detail. Preserve historical material: archive photographs, newspaper clippings, and heritage images can be enhanced for research, publication, or exhibition. Whatever the source, the process is the same — no software installation and no sign-up are required: upload a photo, and the restoration runs automatically, usually ready to download within seconds.
FAQ
Is this tool free?
Yes, completely free with no sign-up and no usage limits.
Which image formats are supported?
PNG, JPG, JPEG and WEBP, up to 40MB per file.
What can it fix?
It automatically removes noise (high ISO / low light), blur, and compression artifacts — making old photos and low-quality images sharper, and auto-detects the degradation type.
Which degradation types are supported?
The model uniformly supports 8 types: super-resolution, denoising, deblurring, derainstreak, deraindrop, dehazing, desnowing, and low-light enhancement — a single model auto-detects and restores, no manual type selection needed.
What is the output format?
A restored PNG image, ready to download.
How long does processing take?
Usually 5-30 seconds per image; large images are auto-shrunk to a reasonable size first.
Is my file stored?
No. Images are automatically deleted within 24 hours after processing.
Can I close the page after uploading?
Yes. Your task is saved in this browser automatically — come back within 24 hours to see the results (only the latest task is kept; do not use private browsing or clear your browser cache).
What technology does it use?
A deep-learning generative model that rebuilds image details directly at high resolution, automatically adapting to different kinds of degradation — scratches, noise, blur and compression artifacts — with no manual tuning needed.
How good is the restoration?
Denoising and deblurring are very effective (measured +5–11 dB PSNR on noisy photos). Note that AI plausibly generates high-frequency detail, so extremely degraded images may differ from the original appearance; low-light images get clarity enhancement rather than brightness restoration.