V2 generate noise instead of images.

[ozeron]

First of all, thank you for your work, you are amazing. 🙇‍♂️ 🙇‍♂️ 🙇‍♂️

I tried from scratch V2 notebook a few times and it consistently generate only noise, I tried a few prompts different variations of sampler, number of training. It generates noisy images, which are far from what I prompt.

I tried a picture of person with negative prompt/and without (((stock photo))), ((((ugly)))), (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), grayscale, black and white, (((bad composition))), ((((stock photo))))

Tried various sampling steps:

• changing sampling steps
• changing steps
• playing with sampler
• prompts
• CFG scale
• picked 512 model and generated 512 pix images

Any suggestions?

P.S. wiped everything. Remove sd folder and got same results:

[TheLastBen]

Hi, did you delete/rename the folder sd recently ? if you didn't you should, and always use the colab from the link to avoid being out of date : https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb

[Topzie]

We've all been there, just change the resolution to the trained model 768 x 768 (or higher)

[Ted-Teddy]

I had the same problem with the V2 colab version for free colab users. What worked for me is: Select under the point "Start stable-diffusion -> Model_Version:" the model V2-768 instead of the V2-512 (with the V2-512 comes only noise, no matter what you set in the Web-UI). After you have executed the point and started the web UI, set the resolution to 768x768 and everything works normally without noise.

[acjbizar]

I had the same problem with the V2 colab version for free colab users. What worked for me is: Select under the point "Start stable-diffusion -> Model_Version:" the model V2-768 instead of the V2-512 (with the V2-512 comes only noise, no matter what you set in the Web-UI). After you have executed the point and started the web UI, set the resolution to 768x768 and everything works normally without noise.

That fixed it for me, thanks! However, I still do get noise with some settings in the Web UI. Not sure which specifically are triggering this.

[orshi]

Have u fixed the issue?

[RRooyyCChheenn]

i started to change the config file from v2-inference.yaml to v2-inference-v.yaml , then fixed this problem. hope it can help you