Accuracy of the diffusion generative AI model proposed in [11]; the quantity is the Truth Negative rate for the ALASKA2 dataset [10, 9] of genuine photographs and true positive rates for all other generative-AI image datasets.
| Source | Accuracy |
|---|---|
| ALASKA2 real photographs | 97.80% |
| Stable Diffusion XL Turbo | 85.01% |
| — with Consistency Decoder VAE | 7.79% |
| Stable Diffusion v3 | 96.07% |
| — with 2 diffusion steps | 16.93% |
| Wuerstchen | 85.01% |
| — with DDIM noise scheduler | 96.20% |
| Kandinsky v2.2 | 93.85% |
| — with 64 diffusion steps | 7.79% |
| FLUX.1 Schnell | 53.81% |
| — with Finetuning #1 | 63.78% |
| — with Finetuning #2 | 47.76% |
| Source | Accuracy |
|---|---|
| ALASKA2 real photographs | 97.80% |
| Stable Diffusion XL Turbo | 85.01% |
| — with Consistency Decoder VAE | 7.79% |
| Stable Diffusion v3 | 96.07% |
| — with 2 diffusion steps | 16.93% |
| Wuerstchen | 85.01% |
| — with DDIM noise scheduler | 96.20% |
| Kandinsky v2.2 | 93.85% |
| — with 64 diffusion steps | 7.79% |
| FLUX.1 Schnell | 53.81% |
| — with Finetuning #1 | 63.78% |
| — with Finetuning #2 | 47.76% |
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.