This is not an AI detector, and a score is not evidence about a person It measures periodic structure in an image's frequency spectrum. On this tool's own bundled templates, a real photograph and the modern diffusion outputs all score under 2x while a flatbed scan scores about 23x, so the highest score in that set belongs to the one image that is definitely not AI. Open for the full scope limits.
What this tool actually measures
One number. The Grid Prominence score is the ratio of peak linear power in a ±2° window at the cardinal angles of the azimuthal profile against the surrounding ±15° baseline, taken from the FFT of a Hanning-windowed, Laplacian-filtered blue channel.
That is a measure of how periodic an image's fine texture is. Nothing anywhere in that chain is specific to AI. The score exists because some older generators left a periodic upsampling signature, not because periodicity means synthesis.
What a score does not tell you
- Whether the image was generated by AI. This is the headline limit and it is not a hedge. Modern diffusion models (SDXL, DALL-E 3, Midjourney v6 and later) largely eliminated the checkerboard artifact this method looks for, and they measure in the same 1.1x to 1.4x band as real photographs.
- Which model produced an image, when, at what settings, or whether it was edited afterwards.
- Whether an image was manipulated, composited, retouched, or reused out of context. This is not a manipulation detector.
- Anything read from provenance. C2PA and Content Credentials, EXIF, IPTC and invisible watermarks are not examined at all, and those are the mechanisms actually designed to answer this question.
- Anything about colour. Only the blue channel is analysed, and the image is downsampled to a 4096px cap before the transform.
A high score means periodic structure. It does not mean AI, and this tool has never been able to say otherwise.
Ordinary things that raise the score
Every one of these is a false positive waiting to happen, and each is more common than AI generation:
- Flatbed and drum scans. Scanner halftone and demosaic patterns are genuinely periodic; the bundled scan template scores about 23x.
- Any resize, upscale, or screenshot, including one taken of a real photograph on a screen.
- JPEG block structure, heavy compression, and repeated re-encoding.
- Print halftone, moiré, and photographs of printed material.
- Sensor pattern noise and demosaic artifacts from the camera itself.
- Subject matter that is simply periodic: fabric, brick, tiling, grilles, architecture.
Never use this for
- Academic integrity or misconduct proceedings.
- Employment, contracting, admissions, grading, or platform enforcement decisions.
- Journalism, fact-checking, or publishing a claim that an image is synthetic.
- Legal, forensic, or insurance determinations, or anything entering a record about somebody.
- Accusing any person of anything, in public or in private.
A score is a statement about an image's spectrum. It is not a statement about the person who made it, and it will not survive being treated as one.
What is checked, and what went wrong before
The transform itself is verified: the JavaScript FFT is tested against analytic references, the Rust and WASM engines are held to parity with it, and the scoring is exercised on synthetic azimuthal profiles where the answer is known. Seeded noise measures about 1.25x and 3px stripes about 85x. That is a working signal-processing pipeline, and it is not the same claim as detecting AI.
Worth knowing, because it shows how quietly this can go wrong: until July 2026 the one-pixel border the Laplacian kernel cannot reach was left unfiltered, which stamped an artificial cross onto the spectrum at exactly the cardinal angles the score measures. Flat grey scored around 19,800x. Every number the tool produced before that fix was wrong, and it looked entirely plausible while it was.
If you actually need to know
Check provenance rather than pixels: C2PA Content Credentials, the original file and its metadata, the capture device, and the chain of custody from whoever made it. Where the answer carries a consequence for someone, that is a question for a qualified digital forensics examiner, not for a heuristic in a browser tab.
Forensic Dashboard
No Image Loaded
1. Log Visuals vs. Linear Truth: The images use Log scaling so you can see them. Internally, we score using Linear power to catch "invisible" grid spikes that are mathematically massive. This is standard practice in signal processing: log for visibility, linear for accuracy.
2. Why Laplacian? We run a High-Pass filter (3×3 Laplacian kernel) to strip image content. The filter sums to zero, guaranteeing removal of the DC component (average brightness). We're looking for the manufacturing defects of the generator, not the picture itself.
3. Artifact Identification:
- Star Field / Grid: Periodic grid indicates upsampling artifacts from transposed convolutions (checkerboard effect). Most common in GANs (StyleGAN1/2, ProGAN) and early diffusion models.
- Solid Halo / Ring: Indicates heavy JPEG compression, aggressive post-processing, or Gaussian blurring.
- Natural Falloff: Real photographs typically exhibit chaotic, non-periodic frequency distributions without geometric patterns.
4. Model Effectiveness:
- Best Results: GAN-generated faces (StyleGAN, ProGAN, BigGAN), AI upscalers, and images with known upsampling artifacts.
- Limited Results: Modern diffusion models (SDXL, DALL-E 3, Midjourney v6+) have largely eliminated checkerboard artifacts through alternative upsampling methods.
- Not Reliable For: Determining whether an image is AI-generated, at all. Modern diffusion output scores in the same band as real photographs, and the highest-scoring bundled template is a flatbed scan. Read the scope disclaimer at the top of the page before drawing any conclusion from a score.
Technical Methodology & Signal Processing
1. The Pipeline
Unlike simple metadata checkers, this tool performs frequency domain analysis. The image is processed in five stages:
Input Image -> Blue Channel Extraction -> Laplacian High-Pass -> Hanning Window -> FFT
2. Channel Selection (Blue Channel)
We default to the Blue Channel. In digital sensors using a Bayer Filter, color sampling is non-uniform: 50% green, 25% red, 25% blue. The blue channel has the lowest sampling density and typically the lowest signal-to-noise ratio. Compression artifacts and generation errors are most prominent in this channel, making it the "canary in the coal mine" for forensic analysis.
3. Edge Detection (Laplacian High-Pass)
A 3×3 Laplacian kernel acts as a second-order derivative filter. Because the kernel sums to zero, it guarantees removal of the DC component (average pixel intensity), leaving only high-frequency content. This effectively removes the "picture" (faces, objects, scenes) and reveals the underlying "texture" (pixel-level relationships, compression artifacts, generation defects). The one-pixel border, which the kernel cannot reach, is zeroed after filtering; leaving raw intensities there would paint a bright frame around the zero-centered interior and stamp a large artificial cross onto the spectrum at exactly the cardinal angles the grid score measures.
4. Windowing (Hanning)
Before FFT, a Hanning window is optionally applied to reduce spectral leakage. Image boundaries create artificial discontinuities (sharp edges) that would appear as spurious high-frequency content. The window tapers these edges smoothly to zero, minimizing artifacts at the cost of slight frequency resolution reduction. Disable this to see edge artifacts more clearly.
5. Frequency Domain (FFT)
We use a Fast Fourier Transform to convert spatial data (pixels) into frequency data (sine wave components). The output is centered (fftshift) so low frequencies appear at the center.
Center: Low frequencies (gradual gradients, smooth regions)
Edges: High frequencies (sharp edges, noise, fine detail)
Transposed Convolutions & Checkerboard Artifacts:
Many GANs and older generative models use transposed convolutions (also called deconvolution) for upsampling. This operation can create checkerboard patterns in the spatial domain that are subtle enough to be invisible to the eye, but mathematically regular. In the frequency domain, these artifacts manifest as a periodic grid or star pattern at regular intervals, corresponding to the upsampling factor.
6. FFT Engine: WASM vs JavaScript
The tool supports two FFT implementations with automatic fallback:
WASM (Rust) - Preferred:
• Handles any image dimensions (not limited to powers of 2), so no resampling at native size
• 3-10x faster through native compiled code and SIMD optimizations (rustfft)
• Returns centered output directly (no post-shift needed)
JavaScript - Fallback:
• Custom Cooley-Tukey implementation (requires power-of-2 dimensions)
• Resamples the image to a power-of-2 square (e.g., 3151×4096 to 4096×4096), which distorts aspect ratio and adds interpolation artifacts; prefer WASM for forensic fidelity
• Pure JS, works in any browser without WASM support
• Separable 2D FFT: rows then columns (O(n² log n) complexity)
Both engines cap analysis at 4096px on the longest side to bound memory use; larger images are downscaled proportionally and the header notes the analyzed size.
7. Grid Prominence Score
The tool calculates a "Grid Prominence Score" by comparing peak linear power in a ±2° window at cardinal angles (0°, 90°, 180°, 270°) against the surrounding ±15° baseline of the azimuthal profile, always on linear power regardless of the Visual Log Scale toggle. Higher ratios indicate stronger periodic artifacts. The score and a qualitative band are shown in the sidebar. On the bundled templates the real photo and the modern diffusion outputs all sit below 2x, while the flatbed scan scores around 23x: scanner halftone and demosaic patterns are genuinely periodic, so a high score means "periodic structure", not "AI". Treat it as one heuristic among many, never a verdict.
8. References & Further Reading
• Detecting and Simulating Artifacts in GAN Fake Images (Zhang et al., 2019)
• Checkerboard artifacts free convolutional neural networks (Sugawara, 2019)
• Discrete Fourier Transform in Unmasking Deepfake Images (MDPI, 2024)
• On the Frequency Bias of Generative Models (Schwarz et al., NeurIPS 2021)