search

Zheng-Chong / CatVTON

CatVTON is a simple and efficient virtual try-on diffusion model with 1) Lightweight Network (899.06M parameters totally), 2) Parameter-Efficient Training (49.57M parameters trainable) and 3) Simplified Inference (< 8G VRAM for 1024X768 resolution).
Other
887 stars 104 forks source link

Potential NSFW content was detected in one or more images. #22

Closed yatoubusha closed 3 months ago

yatoubusha commented 3 months ago

when i run app.py, i will get this error: A black image will be returned instead. Try again with a different prompt and/or seed. image

Zheng-Chong commented 3 months ago

Since the training of the base model SD1.5 contains NSFW data, we added SafetyChecker to the code to prevent potential NSFW from being displayed. You can skip it in the code by setting skip_safety_check=True, but we don't recommend it.

    pipeline = CatVTONPipeline(
        ...
        skip_safety_check=True
    )
yatoubusha commented 3 months ago

Since the training of the base model SD1.5 contains NSFW data, we added SafetyChecker to the code to prevent potential NSFW from being displayed. You can skip it in the code by setting skip_safety_check=True, but we don't recommend it.

    pipeline = CatVTONPipeline(
        ...
        skip_safety_check=True
    )

Thanks, but why is the frequency of this problem so high? I have tried changing the seed multiple times but it still works like this. Additionally, the official demo webpage cannot be opened. Can you please restart it? Why didn't I encounter this problem frequently when I tried it on the official demo webpage before?

Zheng-Chong commented 3 months ago

We used SD's safetychecker directly, the high frequency may be because its settings need to filter potential NSFW as much as possible. The demo was offline on 8.1 and we have restarted it on 8.4. The safetychecker in the online demo has no difference from the one in the code.

yatoubusha commented 3 months ago

We used SD's safetychecker directly, the high frequency may be because its settings need to filter potential NSFW as much as possible. The demo was offline on 8.1 and we have restarted it on 8.4. The safetychecker in the online demo has no difference from the one in the code.

Thank you very much for your reply. After trying to skip, I was able to generate the image normally, and it is not from nsfw. Perhaps the filtering criteria were set too strictly. The recovery of the web version demo is really good news. Thank you for your open source work.