Compare image similarity locally
Compare JPG, PNG and WebP images with local image embeddings. Review similarity scores and groups as suggestions, and inspect originals before deleting photos.
About Image Similarity
Compare images using CLIP image embeddings and review pairwise similarity scores and suggested groups. Processing runs on your device; model and runtime assets download when needed. Scores are suggestions for manual review, not proof that two photos are duplicates. Inspect originals before deleting photos.
- Category
- inspect
- Input
- Accepts: image/jpeg, image/png or image/webp.
- Output
- Outputs: application/json.
- Cost
- Free, runs in your browser
- Memory
- medium
- Install group
- image-ai
Common uses
- Review a small group of images and inspect pairs with high similarity scores.
- Download scores and suggested groups as JSON while retaining your original files.
- Check suspected matches with pixel differences or file checksums when you need stronger evidence.
Frequently asked questions
What does the score mean?
The score is cosine similarity between CLIP image embeddings. A higher score indicates more similar embeddings, but does not prove that the files or pictured objects are identical.
Is this the same as a pixel diff?
No. This tool compares image embeddings. Use Image Diff for a pixel-level comparison of supported same-size images, or Hash to check whether file bytes are identical.
Do the images leave my device?
Image processing runs locally in your browser without uploading your image contents. Model and runtime assets download when needed.
Which formats are supported?
The tool accepts JPEG, PNG and WebP and requires at least two images.
Do the images have to be the same size?
No. Images are decoded and prepared for the embedding model. Different dimensions are accepted, but scores still require manual review.
Will it match two photos of the same object taken differently?
A score does not establish that two photos show the same object. This tool suggests pairs to inspect; it does not automatically confirm duplicates or delete files.
Keywords
- similarity
- duplicate
- compare
- clip
- perceptual
- near-duplicate
- embedding