# Forensic Detection on FakeClass FakeBench (qualitative setting)

Source: [Forensic Detection on FakeClass FakeBench (qualitative setting)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-17T07:00:51.206Z

Evaluation of Forensic Detection performance on the qualitative setting subset of the FakeClass FakeBench dataset, reporting metrics related to specific generative models (e.g., proGAN, SD, DALL·E3) and overall performance.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Forensic Detection |
| Dataset (including any stated split) | FakeClass FakeBench (qualitative setting) |
| Metric | Detection Rate (Fake) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Detection Rate (Fake)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(Fake)>): Higher is better; unit: PERCENT.

- [Detection Rate (Real)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(Real)>): Higher is better; unit: PERCENT.

- [Performance (What)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Performance%20(What)>): Higher is better; unit: PERCENT.

- [Accuracy (Yes/No)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Accuracy%20(Yes%2FNo)>): Higher is better; unit: PERCENT.

- [Detection Rate (proGAN)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(proGAN)>): Higher is better; unit: PERCENT.

- [Detection Rate (styleGAN)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(styleGAN)>): Higher is better; unit: PERCENT.

- [Detection Rate (CogView)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(CogView)>): Higher is better; unit: PERCENT.

- [Detection Rate (FuseDream)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(FuseDream)>): Higher is better; unit: PERCENT.

- [Detection Rate (VQDM)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(VQDM)>): Higher is better; unit: PERCENT.

- [Detection Rate (Glide)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(Glide)>): Higher is better; unit: PERCENT.

- [Detection Rate (SD)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(SD)>): Higher is better; unit: PERCENT.

- [Detection Rate (DALL·E2)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(DALL%C2%B7E2)>): Higher is better; unit: PERCENT.

- [Detection Rate (DALL·E3)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(DALL%C2%B7E3)>): Higher is better; unit: PERCENT.

- [Detection Rate (MJ)](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Detection%20Rate%20(MJ)>): Higher is better; unit: PERCENT.

- [Overall Detection Rate](<https://www.sota2.com/research/sota/forensic-detection-on-fakeclass-fakebench-qualitative-setting.md?metric=Overall%20Detection%20Rate>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Detection Rate (Fake) | Result date | Report paper |
| --- | --- | --- | --- |
| FakeScope | 99.5 | Not recorded | 68fa77af1e209131cea74b68 |
| FakeScope | 98.67 | Not recorded | 68fa77af1e209131cea74b68 |
| Human (Best) | 97 | Not recorded | 68fa77af1e209131cea74b68 |
| Human (Overall) | 76.91 | Not recorded | 68fa77af1e209131cea74b68 |
| InstructBLIP | 67.8 | Not recorded | 68fa77af1e209131cea74b68 |
| GPT-4V | 59.87 | Not recorded | 68fa77af1e209131cea74b68 |
| mPLUG-Owl2 | 49.6 | Not recorded | 68fa77af1e209131cea74b68 |
| Human (Worst) | 43 | Not recorded | 68fa77af1e209131cea74b68 |
| Random guess | 42.27 | Not recorded | 68fa77af1e209131cea74b68 |
| LLaVA-v1.5 | 38 | Not recorded | 68fa77af1e209131cea74b68 |
| GeminiPro | 35.83 | Not recorded | 68fa77af1e209131cea74b68 |
| InternLM-XC.2-vl | 32.17 | Not recorded | 68fa77af1e209131cea74b68 |
| Qwen-VL | 28.57 | Not recorded | 68fa77af1e209131cea74b68 |
| IDEFICS-Instruct | 24.97 | Not recorded | 68fa77af1e209131cea74b68 |
| Claude3 Sonnet | 12 | Not recorded | 68fa77af1e209131cea74b68 |

### FakeScope

Detection Rate (Fake): 99.5. Result ID: 69b8fc1b47f34ce924a0ed6a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

Recorded evaluation configuration: {"Backbone":"mPLUG-Owl2"}. This record does not assert independent reproduction.

### FakeScope

Detection Rate (Fake): 98.67. Result ID: 69b8fc1b47f34ce924a0ed6b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

Recorded evaluation configuration: {"Backbone":"LLaVA-v1.5"}. This record does not assert independent reproduction.

### Human (Best)

Detection Rate (Fake): 97. Result ID: 69b8fc1b47f34ce924a0ed5e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Human (Overall)

Detection Rate (Fake): 76.91. Result ID: 69b8fc1b47f34ce924a0ed60.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### InstructBLIP

Detection Rate (Fake): 67.8. Result ID: 69b8fc1b47f34ce924a0ed65.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6fb71e209131cea47021.md>)

[Implementation](<https://github.com/salesforce/LAVIS/tree/main/projects/instructblip>)

Recorded evaluation configuration: {"Backbone":"Vicuna-7B"}. This record does not assert independent reproduction.

### GPT-4V

Detection Rate (Fake): 59.87. Result ID: 69b8fc1b47f34ce924a0ed61.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa727a1e209131cea56a1e.md>)

[Implementation](<https://github.com/openai/evals>)

Recorded evaluation configuration: {"Type":"Proprietary, teacher"}. This record does not assert independent reproduction.

### mPLUG-Owl2

Detection Rate (Fake): 49.6. Result ID: 69b8fc1b47f34ce924a0ed69.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71201e209131cea4f575.md>)

[Implementation](<https://github.com/X-PLUG/mPLUG-Owl/tree/main/mPLUG-Owl2>)

Recorded evaluation configuration: {"Backbone":"LaMA2-7B","Status":"baseline"}. This record does not assert independent reproduction.

### Human (Worst)

Detection Rate (Fake): 43. Result ID: 69b8fc1b47f34ce924a0ed5f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Random guess

Detection Rate (Fake): 42.27. Result ID: 69b8fc1b47f34ce924a0ed5d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### LLaVA-v1.5

Detection Rate (Fake): 38. Result ID: 69b8fc1b47f34ce924a0ed68.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa717d1e209131cea51970.md>)

[Implementation](<https://llava-vl.github.io/>)

Recorded evaluation configuration: {"Backbone":"Vicuna-7B","Status":"baseline"}. This record does not assert independent reproduction.

### GeminiPro

Detection Rate (Fake): 35.83. Result ID: 69b8fc1b47f34ce924a0ed62.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78731e209131cea78622.md>)

Recorded evaluation configuration: {"Type":"Proprietary"}. This record does not assert independent reproduction.

### InternLM-XC.2-vl

Detection Rate (Fake): 32.17. Result ID: 69b8fc1b47f34ce924a0ed64.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72021e209131cea541e3.md>)

[Implementation](<https://github.com/InternLM/InternLM-XComposer>)

Recorded evaluation configuration: {"Backbone":"InternLM2-7B"}. This record does not assert independent reproduction.

### Qwen-VL

Detection Rate (Fake): 28.57. Result ID: 69b8fc1b47f34ce924a0ed66.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70d41e209131cea4d774.md>)

[Implementation](<https://github.com/QwenLM/Qwen-VL>)

Recorded evaluation configuration: {"Backbone":"Qwen-7B"}. This record does not assert independent reproduction.

### IDEFICS-Instruct

Detection Rate (Fake): 24.97. Result ID: 69b8fc1b47f34ce924a0ed67.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70581e209131cea4a712.md>)

[Implementation](<https://github.com/huggingface/OBELICS>)

Recorded evaluation configuration: {"Backbone":"LLaMA-7B"}. This record does not assert independent reproduction.

### Claude3 Sonnet

Detection Rate (Fake): 12. Result ID: 69b8fc1b47f34ce924a0ed63.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77af1e209131cea74b68.md>)

Recorded evaluation configuration: {"Type":"Proprietary"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
