The Search for Statistical Indistinguishability Between Real and Fake Images

Mohaimin Rafi; Otavio Moraes Rosa; Anshuman Chhabra; Utkarsh Ojha

cs.CVOct 28, 2026

The Search for Statistical Indistinguishability Between Real and Fake Images

Authors: Mohaimin Rafi, Otavio Moraes Rosa, Anshuman Chhabra, Utkarsh Ojha

Organizations: University of South Florida

Abstract

Images produced by generative models appear strikingly similar to real ones. In this work, we study whether perfection has been reached by generated content in terms of matching the properties of real data. We train deep neural networks to separate real data from fake and observe their test accuracies. Only a chance performance has any scope of proving statistical indistinguishability. However, since generative models are known to sometimes introduce small errors, e.g., low level checkerboard artifacts, we create progressive abstractions of the original real/fake images to potentially bring them to a space where those artifacts get wiped away. These include operations such as blurring, downsizing, cropping, converting RGB images into edge/semantic maps, etc. To our surprise, almost none of the abstractions are able to make fake data indistinguishable from real. The only cases when neural network based classifiers fail to separate the two categories is when the data is abstracted to such an extent that the original content seems mostly lost to the human eye, e.g., downsizing a $256 \times 256$ RGB image of human face to $4 \times 4$ resolution. We observe this trend across four different datasets of varying complexity (MNIST, CelebA, FFHQ and GenImage) and across various types of generative models (GANs, VAEs, diffusion models). This shows that the nature of fake artifacts extends all the way from low level to very high level in images. The simple implication is that fake images are much more different from real than we are used to think.

Figures & tables

Figure 1: The first column corresponds to a classifier being trained to distinguish clean real (top) and fake images (bottom), achieving a high test accuracy because of the presence of artifacts. In each subsequent column, we transform the original images with the hope of masking away the artifact signal; e.g., downsizing (DS) to $16\times 16$ (6th column). When new classifiers are trained for each column, the accuracy still remains high. For true statistical indistinguishability ( $\sim 50\%$ ; last two columns), we need to abstract away most of the image content by, e.g., downsizing to $4\times 4$ resolution.

Figure 2: Left : even when visually mild abstractions such as blur, Gaussian noise, uniform noise, quantization, or brightness shifts are applied, binary neural classifiers can still separate real and fake MNIST samples with high test accuracy. Right : toy examples contrasting the desired i.i.d. case, $\mathcal{R},\mathcal{F}\sim S$ , with a non-i.i.d. shifted fake distribution, $\mathcal{F}=\mathcal{R}+\delta$ , where the two sample sets are visually close but statistically structured differently.

Figure 3: Motivation for image abstractions. Left: fake FFHQ images may contain localized artifacts, such as unnatural skin, that disappear after mapping the image to a higher-level semantic segmentation representation. Right: CelebA examples show a spatial abstraction, where downsampling removes fine image details while preserving coarse facial structure. We apply such abstractions to both real and fake images and test whether classifiers can still distinguish $f(\mathcal{R})$ from $f(\mathcal{F})$ .

Figure 4: MNIST real-vs-fake detection under image abstractions. Columns show clean and corrupted versions of matched real and ACGAN-generated digits, while the heatmap reports the best test accuracy across DenseNet-Small, Mini-Xception, and ResNet-Small. Real and fake images remain separable through many abstractions and approach chance only when the digit content is heavily destroyed. Please see Table 6 for corruption configurations.

Figure 5: FFHQ real (top) and StyleGAN2, VD VAE, LDM-generated faces under various data abstractions. Even after abstracting away enough information from real and fake images, e.g., by converting them into semantic mask, we do not observe statistical indistinguishability, as a classifiers trained from scratch are able to obtain accuracies way above $50\%$ in Table 1 .

Model	Downsizing				Cropping				Blurring
Model	$64{\times}64$	$16{\times}16$	$4{\times}4$	$1{\times}1$	$64{\times}64$	$16{\times}16$	$4{\times}4$	$1{\times}1$	$\sigma{=}5$	$\sigma{=}10$	$\sigma{=}20$	$\sigma{=}40$
StyleGAN2	99.8%	86.2%	58.4%	52.6%	100%	99.8%	96.3%	54.8%	98.0%	93.0%	74.0%	63.0%
VD VAE	98.9%	91.9%	69.3%	57.0%	100%	99.8%	85.9%	57.8%	95.8%	93.2%	83.0%	73.6%
LDM	99.5%	73.7%	58.1%	54.7%	100%	99.6%	82.5%	55.5%	93.0%	79.6%	64.8%	59.1%

Table 1: FFHQ real-vs-fake test accuracy of the best performing classifier under various data abstractions for StyleGAN2, VD VAE and LDM. Detection remains high for many spatially degraded inputs and approaches chance only under the strongest abstractions, such as

1\times 1

spatial collapse.

	Edge	Mask	Keypoint	Eye	Nose	Mouth	L-eye	R-brow	R-ear
StyleGAN2	91.7%	76.1%	78.3%	99.8%	99.9%	99.9%	99.2%	98.9%	99.0%
VD VAE	95.0%	70.3%	76.5%	99.9%	99.9%	99.9%	99.9%	99.9%	99.9%
LDM	99.8%	67.8%	75.0%	99.9%	100.0%	100.0%	99.9%	99.9%	99.9%

Table 2: FFHQ real-vs-fake test accuracy of best performing classifier for StyleGAN2, VD VAE and LDM across various data abstractions (first 3 columns) and cropped face parts (last 6 columns).

Figure 6: Real (top) and fake images (bottom three) from GenImage dataset under different spatial and blur abstractions. The original images are of $256\times 256$ resolution. Even after abstracting the information content to a great extent by downsizing, blurring or compressing, the resulting real and fake images are still distinguishable by a classifier trained from scratch, as shown in Table 3 .

Model	Downsizing			Cropping			Blurring			JPEG
Model	$16\times 16$	$4\times 4$	$1\times 1$	$16\times 16$	$4\times 4$	$1\times 1$	$\sigma=10$	$\sigma=20$	$\sigma=40$	50	25	5
Wukong	90.9%	81.1%	66.7%	97.5%	66.2%	61.9%	91.4%	87.2%	83.9%	96.6%	96.0%	92.6%
SD v1.4	91.1%	80.9%	68.6%	76.1%	61.5%	58.7%	91.1%	87.5%	82.4%	97.5%	94.8%	75.9%
SD v1.5	89.7%	80.2%	67.8%	75.6%	61.2%	58.4%	89.9%	86.0%	81.2%	96.9%	95.8%	91.1%
Midjourney	78.4%	69.1%	58.7%	93.2%	73.5%	57.4%	81.2%	74.7%	70.3%	92.7%	90.2%	84.3%

Table 3: GenImage real-vs-fake test accuracy under downsampling, cropping, blurring and JPEG abstractions. Detection remains well above chance for most abstractions and approaches chance only under extreme spatial collapse, such as

1\times 1

downsampling or cropping.

Appendix figures & tables14 assets

Supplementary material from the paper’s appendix.

Appendix

Resource	Description	Usage
Compute environment	Institutional HPC cluster managed with SLURM.	On-premise HPC
Accelerators	NVIDIA GPU nodes scheduled from the pool, including A100, A40, Titan RTX, Titan X, Titan V, and GTX 1080 Ti GPUs.	1 GPU per run
CPU and memory	Standard training allocation used by the reported experiments.	4 CPU cores, 16 GB RAM
Storage	Cluster data filesystem used for datasets, checkpoints, generated samples, logs, and analysis outputs.	personal env
Software environment	Conda environment with PyTorch and torchvision; jobs launched through SLURM scripts.	Python/PyTorch

Appendix

Table 4: Compute environment used for the experiments.

Model	Architecture	Parameters
ResNet-Small	$3{\times}3$ convolution stem with $16$ channels, followed by three residual stages: two $16$ -channel residual blocks, two $32$ -channel residual blocks with stride- $2$ downsampling in the first block, and two $64$ -channel residual blocks with stride- $2$ downsampling in the first block. Each residual block uses two $3{\times}3$ convolutions with batch normalization and ReLU, plus an identity or $1{\times}1$ projection shortcut. Final adaptive average pooling and a linear sigmoid head.	$174{,}945$
DenseNet-Small	DenseNet-BC style network with growth rate $k=12$ , initial $3{\times}3$ convolution to $24$ channels, three dense blocks with four dense layers per block, and transition layers between blocks. Each dense layer uses BN–ReLU– $1{\times}1$ convolution followed by BN–ReLU– $3{\times}3$ convolution, then concatenates features with the input. Final batch normalization, adaptive average pooling, and a linear sigmoid head.	$101{,}479$
Mini-Xception	Compact Xception-style network. Entry block: $1\to 8$ channels via a $3{\times}3$ convolution, batch normalization, and ReLU. Three residual depthwise-separable blocks with channel sizes $16$ , $32$ , and $64$ ; the first two downsample by a factor of $2$ . Final adaptive average pooling and a linear sigmoid head.	$13{,}089$

Appendix

Table 5: Binary classifier architectures used for the MNIST real-vs-fake abstraction experiments.

Figure 7: MNIST real-vs-fake detection under image abstractions. Columns show clean and corrupted versions of matched real and ACGAN-generated digits, while the heatmap reports test accuracy for DenseNet-Small (DNet), Mini-Xception (XCP), and ResNet-Small (RNet). Real and fake images remain separable through many abstractions and approach chance only when the digit content is heavily destroyed. Please see Table - 6 for corruption configurations.

#	Operation
0	clean
1	blur $\sigma{=}3.0$
2	gauss $\sigma{=}0.5$
3	quant $L{=}2$
4	unif $a{=}0.5$
5	brightness $b{=}+0.95$

Appendix

Table 6: The 25 abstraction configurations used for the main MNIST experiment.

Name	Optimizer	Params	Description
Baseline CNN	Adadelta 1.0	7,392,449	The model uses two convolutional layers, mapping $3$ channels to $32$ and then $32$ to $64$ , followed by max pooling and dropout with probability $0.25$ . The flattened features pass through a $128$ -unit fully connected layer, dropout with probability $0.5$ , and a final one-unit output layer.
CNN+BatchNorm	Adam $10^{-3}$	7,392,641	Baseline with BatchNorm after each conv; dropout removed.
AC-GAN Disc.	Adam $10^{-3}$	99,937	4 stride-2 conv layers, LeakyReLU(0.2), BatchNorm, Dropout(0.25). Mirrors AC-GAN discriminator.
ResNet-Small	Adam $10^{-3}$	175,233	3 $\times$ 3 stem; 3 residual blocks (16 $\to$ 32 $\to$ 64 ch, stride-2); AdaptiveAvgPool; FC(64 $\to$ 1).
Constrained CNN	Adam $10^{-3}$	189,505	Zero-sum constrained first conv (5 $\times$ 5); 2 conv layers; AdaptiveAvgPool(4); FC(1024 $\to$ 128 $\to$ 1).
DenseNet-Small	Adam $10^{-3}$	101,911	3 dense blocks, growth rate $k=12$ ; AdaptiveAvgPool.

Appendix

Table 7: Architectures used in the CelebA experiments.

Figure 8: PCA results for CelebA/DcGAN configuration

Figure 9: CelebA real and DCGAN-generated faces under spatial and blur abstractions.

Model	Downsizing				Cropping				Blurring
Model	$16\times 16$	$4\times 4$	$2\times 2$	$1\times 1$	$10\times 10$	$4\times 4$	$2\times 2$	$1\times 1$	$\sigma=6$	$\sigma=10$	$\sigma=20$	$\sigma=30$
Constrained CNN	99.7%	90.2%	68.9%	53.9%	100.0%	99.9%	94.6%	57.1%	97.3%	91.7%	93.4%	96.2%
DenseNet	99.5%	89.0%	68.2%	54.0%	100.0%	100.0%	94.6%	57.1%	98.2%	91.0%	90.6%	91.2%
SRNet	99.8%	90.5%	68.4%	54.1%	100.0%	99.9%	94.5%	57.1%	96.4%	84.1%	83.6%	86.5%

Appendix

Table 8: CelebA real-vs-fake test accuracy under downsampling, cropping, and blurring abstractions. Each entry reports binary classification accuracy on held-out real and DCGAN-generated test images. Detection remains well above chance for most abstractions and approaches chance only under extreme spatial collapse, such as

1\times 1

downsampling or cropping.

Name	Optimizer	Params (1ch)	Params (3ch)	Description
ResNet	Adam $10^{-3}$	174,385	174,817	Stem Conv+BN+ReLU; 3 residual stride-2 layers (16 $\to$ 32 $\to$ 64 ch); AdaptiveAvgPool; FC(64 $\to$ 1).
DenseNet	Adam $10^{-3}$	133,481	134,057	Stem Conv(stride=2)+MaxPool; 3 DenseBlocks (growth rate 16, 4 layers each); AdaptiveAvgPool; FC $\to$ Sigmoid.
ViT	Adam $10^{-4}$	12,874,241	12,875,265	patch_size=16, dim=512, depth=6, heads=8, mlp_dim=1024, dropout=0.1.

Appendix

Table 9: Architectures used in the FFHQ experiments. Parameter counts are given for 1-channel (derived representations) and 3-channel (raw RGB) inputs.

Figure 10: FFHQ real and StyleGAN2-generated faces under semantic and local-part abstractions.

Figure 11: FFHQ real and StyleGAN2-generated faces under spatial and blur abstractions.

Model	Downsizing				Cropping				Blurring
Model	$64{\times}64$	$16{\times}16$	$4{\times}4$	$1{\times}1$	$64{\times}64$	$16{\times}16$	$4{\times}4$	$1{\times}1$	$\sigma{=}5$	$\sigma{=}10$	$\sigma{=}20$	$\sigma{=}40$
ResNet	99.1%	67.2%	54.5%	52.6%	100.0%	99.7%	81.7%	54.7%	96.5%	76.9%	57.0%	55.2%
DenseNet	99.5%	77.6%	55.5%	52.5%	100.0%	99.8%	96.3%	54.8%	96.7%	82.3%	66.8%	59.6%
ViT	99.8%	86.2%	58.4%	51.3%	100.0%	99.8%	95.1%	50.5%	98.0%	93.0%	74.0%	63.0%

Appendix

Table 10: FFHQ real-vs-fake test accuracy under downsampling, cropping, and blurring abstractions. Each entry reports binary classification accuracy on held-out real and StyleGAN2-generated test images. Detection remains high for many spatially degraded inputs and approaches chance only under the strongest abstractions, such as

1\times 1

spatial collapse.

	Edge	Mask	Eye	Nose	Mouth	L-eye	R-brow	R-ear
ResNet	82.4%	71.1%	99.69%	99.28%	99.85%	98.97%	98.59%	74.97%
DenseNet	85.9%	76.1%	99.84%	99.90%	99.71%	99.24%	97.19%	86.19%
ViT	91.7%	70.9%	99.83%	99.78%	99.77%	98.34%	98.88%	98.95%

Appendix

Table 11: Real-vs-fake test accuracy on FFHQ vs. StyleGAN2 across derived representations (Edge, Mask) and six cropped face parts.

Figure 12: Real (top) and fake images (bottom three) from GenImage dataset under different spatial and blur abstractions.

Image Generation Evaluation

Image generation evaluation measures the quality, fidelity, diversity, and instruction-following ability of AI-generated images.

See all 116 papers

SimLBR: Learning to Detect Fake Images by Learning to Detect Real Images
Feb 23, 2026Aayush Dhakal, Subash Khanal, Srikumar Sastry +4AI-Generated Image Detection OOD Generalization
Washington University in St. Louis · Oak Ridge National Laboratory
Do CNNs Internally Represent Real and Fake Images Differently? A Hidden-Layer Analysis
Aug 12, 2026Moumita Sen Sarma, Pascal Hitzler, Eugene Y. VassermanNeuron-Level Interpretability Convolutional Neural Networks
Department of Computer Science, Kansas State University, Manhattan, Kansas, USA
A Difference-in-Difference Approach to Detecting AI-Generated Images
Feb 27, 2026Xinyi Qi, Kai Ye, Chengchun Shi +3AI-Generated Image Detection Reconstruction-Based Anomaly Detection
The London School of Economics and Political Science · University of Birmingham

Feb 23, 2026J/K move · Enter open · S save

SimLBR: Learning to Detect Fake Images by Learning to Detect Real Images

Aayush Dhakal, Subash Khanal, Srikumar Sastry, Jacob Arndt+3

Washington University in St. Louis · Oak Ridge National Laboratory

AI-Generated Image Detection OOD Generalization One-Class Classification

Open paper AskPDF arXiv

The rapid advancement of generative models has made the detection of AI-generated images a critical challenge for both research and society. Recent works have shown that most state-of-the-art fake image detection methods overfit to their training data and catastrophically fail when evaluated on curated hard test sets with strong distribution shifts. In this work, we argue that it is more principled to learn a tight decision boundary around the real image distribution and treat the fake category as a sink class. To this end, we propose SimLBR, a simple and efficient framework for fake image detection using Latent Blending Regularization (LBR). Our method significantly improves cross-generator generalization, achieving up to +24.85% accuracy and +69.62% recall on the challenging Chameleon benchmark. SimLBR is also highly efficient, training orders of magnitude faster than existing approaches. Furthermore, we emphasize the need for reliability-oriented evaluation in fake image detection, introducing risk-adjusted metrics and worst-case estimates to better assess model robustness. All code and models will be released on HuggingFace and GitHub.

Abstract

Image Generation Evaluation

Explore similar work

SimLBR: Learning to Detect Fake Images by Learning to Detect Real Images

Do CNNs Internally Represent Real and Fake Images Differently? A Hidden-Layer Analysis

A Difference-in-Difference Approach to Detecting AI-Generated Images