Realism Engine 1.0

This looks gorgeous and might finally be the thing to get me to go from SD 1.5 up to 2.1!

Really a good blend in realism. Any chance to also get a LORA so to easily mix it with other models extending its potential ?

Extremely excited to really dive into this! And just in time for ControlNet for SD 2.1! (ahem, I used your model for the previews, thanks!)

Will review shortly!

Does this model use offset noise in training?

Do you plan to train inpaintning model? It could be useful with Photoshop plugin's.

Can someone share a training guide for such models? I also want to train my own. Would really appreciate!

Can we add this also to Hugging Face?

https://huggingface.co/models

Downloaded safetensors file and yaml, same name, same folder, set image size to 768x768 in automatic1111, but not getting any image output, receiving this error instead:

> NotImplementedError: No operator found for memory_efficient_attention_forward with inputs: query : shape=(1, 9216, 1, 512) (torch.float32) key : shape=(1, 9216, 1, 512) (torch.float32) value : shape=(1, 9216, 1, 512) (torch.float32) attn_bias : <class 'NoneType'> p : 0.0 cutlassF is not supported because: device=mps (supported: {'cuda'}) flshattF is not supported because: device=mps (supported: {'cuda'}) dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) max(query.shape[-1] != value.shape[-1]) > 128 tritonflashattF is not supported because: device=mps (supported: {'cuda'}) dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) max(query.shape[-1] != value.shape[-1]) > 128 triton is not available smallkF is not supported because: device=mps (supported: {'cpu', 'cuda'}) max(query.shape[-1] != value.shape[-1]) > 32 unsupported embed per head: 512

how do you load this .safetensors file into a diffusers from_pretrained ? error seems to imply its missing some files

Sad it cannot be merged with 1.5 models, have to wait until my favourites have 2.1 models versions.

Good work anyways !

Colors are really good, but quality for some reason is quite bad, loads of noise in the images it produces

Sorry for the ignorance, I'm new to ai, so how do you install it in stable diffusion automatic1111

And how to use it

Hi, awesome models!! Quick question.. I am having problems with replicating the same images by using the sample prompts. Currently, I just got 1.5 SD installed, do I need to install 2.1 to achieve the same effect? Thanks in advance!

Browsing through the SD2.1 models, I am amazed by the quality that some of these and especially yours are giving.

Curious though, is there a way to "bake in" the config files into the models, like you can with the VAEs on some SD1.5 models?

Regardless of that, I'll definitely be playing around with some SD2.1based models now, amazing stuff and nice work on these, @razzz . :D

Hi, I wanted to give this model a shot, since I am more interested in creating images of things and places rather than humans. Unfortunately, I cannot find the nfixer nor nrealfixer under the links provided, I'm getting a 404 and couldn't find it through Google. Is there any functioning mirror left? Thanks.

Hi I'm getting the following error and no output. I have no idea what to do:

NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

I'm using a 3070ti, and I tried the suggestions in the error with no success.

Hi,
I get "NansException: A tensor with all NaNs was produced in Unet." error when running this checkpoint on Automatic1111. I'm I doing something wrong or Isn't it compatible?

@razzz for some reason I'm getting this error when I try to use the model.

RuntimeError: The size of tensor a (1024) must match the size of tensor b (768) at non-singleton dimension 1

I tried downloading again but it persists. I'm not using any embeddings or lora, just a simple prompt and the model. Tried different image sizes too (768x768, 768x1024, etc) This error started after I updated torch and xformers.

torch: 2.0.0+cu118

xformers: 0.0.17+b6be33a.d20230315

I'm not even sure those would have some effect on running the model, it's the only thing I changed on my end since I was last able to run it It's a bummer, I really like this model and wish I could run it. Anyway thought I'd get the info out there incase someone else has run into the same issue.

how is this so tiny? isn't it supposed to be 1-2 GB ?

WOW ! Such realistic results dude... M impressed by the results... I am facing issue getting my trained model to give more realistic images, any suggestions on what parameters(lr, steps, num class images) might be of help to me, or anything about the dataset... Thanks in advance 😇😇

The majority of the images in the showcase are made with "_optimal", not "RealismEngine_v10". Is that another model?

Please update the links for nfixer and nrealfixer in the description. They are broken (404)