ImageDiff

Multi-Model AI Image Reproduction Evaluation

Evaluate how well different AI models (Nano Banana, GPT Image, Claude, etc.) reproduce images using multiple perceptual quality metrics.

Test multiple AI models simultaneously (Gemini, GPT-4, Claude)
Compare with advanced metrics: MS-SSIM, SSIMULACRA2, Butteraugli, FLIP, LPIPS, PSNR
Identify which metrics best align with human perception
Determine which model performs best for each metric
Visualize differences with color-coded error maps and distribution analysis

Upload Image

Tip: For best comparison results, upload images with standard aspect ratios (16:9, 4:3, 1:1) or common dimensions (1024×1024, 1920×1080).

Drag & drop an image here or click to browse

Clone Prompt

Prompt Template:

Select AI Models for Comparison

Select one or more models to generate and compare image reproductions

Loading models...

Difference Threshold: 10

Sensitivity for pixel difference detection (0-50)

All Models - Image Comparisons

Generate images with models to see comparisons...

Detailed Metrics Comparison (Current Image)

Click column headers to sort. blue = best, Red = worst.

Model	PSNR ↑	SSIM ↑	SSIMULACRA2 ↑	Butteraugli ↓	FLIP ↓	LPIPS ↓	% Diff ↓
No comparison data yet

Pixel Difference Analysis

Select Model:

Total Pixels:
-

Different:
-

Percent Diff:
-

PSNR:
-

MAE (R): -

MAE (G): -

MAE (B): -

MSE: -

SSIM: -

Structural Similarity (0-1, higher better)

SSIMULACRA2: -

Perceptual quality (0-100, higher better)

Butteraugli: -

Perceptual diff (0-10, lower better)

FLIP: -

Feature-based metric (lower better)

LPIPS: -

Learned Perceptual (lower better)

Error Range	Pixel Count	Percentage	Visual
No comparison data yet