The result is a black square

[x18-1]

I wanna draw something,but when I run the app,the result is a black square。 And no suggestions In Terminal,so I didn't know how to solve the problem.

[amrrs]

If you are getting black square, most likely it's NSFW image. You can try to change your prompt and try!

[lkwq007]

If you are getting black square, most likely it's NSFW image. You can try to change your prompt and try!

@x18-1 The false positive rate of safety checker/NSFW detector is relatively high for this project. So you can also disable the safety checker if you keep getting black square.

[x18-1]

@lkwq007

I try ，but it didn’t work

[lkwq007]

@lkwq007

I try ，but it didn’t work

[x18-1]

@lkwq007 I try ，but it didn’t work

@lkwq007

That's exactly what I did，but didn‘t work

the photo is about app

the photo is about Ternimal

[x18-1]

If you are getting black square, most likely it's NSFW image. You can try to change your prompt and try!

"Potential NSFW content was detected in one or more images. A black image will be returned instead. Try again with a different prompt and/or seed." appears in my terminal。But I have no idea how to do

[Eprise1701e]

Same issue here...Safety unchecked. All the other stable diffusion stuff I've used hates my GTX 1660 and wants precision set to full to show anything but a blank or green square. Wondering if that's the case here, but I'm still a pleb and need guide on how to change this... Not seen anything about it for infinity yet.

[vlsech]

I too have a GTX 1660 Ti. At first I had the same issue in Stable Diffusion from Automatic 1111; however, after I forced FULL PRECISION via CMD LINE with "--opt-split-attention --precision full --no-half " and could generate 16 batches of 4 images IF I add "--medvram --force-enable-xformers. Is there a way to FORCE FULL PRECISION in Infinity?

[lkwq007]

I too have a GTX 1660 Ti. At first I had the same issue in Stable Diffusion from Automatic 1111; however, after I forced FULL PRECISION via CMD LINE with "--opt-split-attention --precision full --no-half " and could generate 16 batches of 4 images IF I add "--medvram --force-enable-xformers. Is there a way to FORCE FULL PRECISION in Infinity?

@vlsech The latest version can run with python app.py --fp32 --lowvram

[vlsech]

I too have a GTX 1660 Ti. At first I had the same issue in Stable Diffusion from Automatic 1111; however, after I forced FULL PRECISION via CMD LINE with "--opt-split-attention --precision full --no-half " and could generate 16 batches of 4 images IF I add "--medvram --force-enable-xformers. Is there a way to FORCE FULL PRECISION in Infinity?

@vlsech The latest version can run with python app.py --fp32 --lowvram

Thanks, just tested.

Am seeing "CUDA out of memory. Tried to allocate 64.00 MiB (GPU 0; 6.00 GiB total capacity; 5.31 GiB already allocated; 0 bytes free; 5.32 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF" Any switches to prevent memory fragmentation? Lastly the NVIDIA, the xformers drivers can provide a 30% speed boost to boot. Thanks.

[jdries3]

I tried python app.py --fp32 --lowvram and got the same kind of error:

CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 4.00 GiB total capacity; 3.41 GiB already allocated; 0 bytes free; 3.46 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

So I tried using PYTORCH_CUDA_ALLOC_CONF to set garbage collection and threshold options. I tried first just setting garbage_collection_threshold:0.6. I upped it to 0.8, neither fixed my problem. I then added max_split_size_mb:128 and tried the values 64, 128, 256. None resolved the problem.

Now I have gotten a Stable Diffusion fork to work. I can make Automatic1111's fork work if I invoke their app with the following parameters:

--lowvram --precision full --no-half --always-batch-cond-uncond --unload-gfpgan --opt-split-attention

Without all of the parameters, I would either run out of VRAM, or get black/green blocks (unrelated to NSFW filters).

[vlsech]

@lkwq007, Infinity looks so promising, hoping you may look at offering more low vram options for us underpowered users such as listed here https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting