Cant get it to generate or find models

[ryan13mt]

Not sure i'm doing anything wrong. I downloaded the windows files, extracted, opened the exe and managed to open the page. I clicked to download the base models shown on the first screen. Downloaded some extra models from civitai screen inside the app.

My top left model chooser is empty, tried restarting app and also restarting machine, same thing.

When clicking enfugue, it initializes and then stops. I can see the GPU memory usage go from 2gb to 8gb which is all i have and it stays like this even after failing to generate so memory is not being released it seems.

Anyone else have such problems? I have auto1111 set up and working ok so this is not my first try at SD and my pc can handle it.

[painebenjamin]

Hi Ryan,

There's definitely some room for me to explain better what you need to do with models. I tried to update the README last night with more information about this, but models aren't just a singular checkpoint, you need to configure models that reference the downloaded checkpoint and other model(s).

For example, after making this model here, The model picker allows me to select "Realism" and is using the cyberrealistic checkpoint.

The reason for this style of configuration over direct file picking is primarily for the maintenance of TensorRT engines. Engines are specific to the checkpoint, size, and other weights (LoRA, TI) that are in it at compile time, so if the user changes any of those variables, the engine will no longer work.

After you initially clicked "Enfugue," about how long did you wait before it looked like nothing was going to happen? The first invocation can take a bit longer than subsequent ones, as it reaches out to Huggingface to pull other details of the models.

[ryan13mt]

To be honest i tried both models that where already installed. The below got the same cuda error, maybe there is no limit being set on how much memory to allocate? I get the same error on auto1111 when i try to generate anything larger than 1536x1536 for example.

{
   "model":"C:\\Users\\User\\.cache\\enfugue\\checkpoint\\v1-5-pruned.ckpt",
   "lora":[
      
   ],
   "inversion":[
      
   ],
   "width":512,
   "height":512,
   "size":512,
   "seed":null,
   "prompt":"orc character",
   "negative_prompt":"",
   "image":null,
   "image_callback_steps":10,
   "nodes":[
      {
         "width":512,
         "height":512,
         "prompt":"orc character",
         "negative_prompt":"",
         "controlnet":null,
         "conditioning_scale":1,
         "strength":0.8,
         "num_inference_steps":50,
         "guidance_scale":7.5,
         "remove_background":false,
         "process_control_image":true,
         "scale_to_model_size":true,
         "image":null,
         "mask":null,
         "control_image":null,
         "children":[
            
         ],
         "bounds":[
            [
               0,
               0
            ],
            [
               512,
               512
            ]
         ]
      }
   ],
   "samples":1,
   "upscale":{
      "method":[
         "esrgan"
      ],
      "amount":1,
      "iterative":false,
      "diffusion":false
   },
   "chunking_size":64,
   "chunking_blur":64,
   "build_tensorrt":false
}

[painebenjamin]

Okay - thanks a lot for all of the information. I'm going to have to do some more digging, I'll let you know what I come up with.

Thanks again!

[ryan13mt]

I'll keep a look at the repo for more updates in the future. Will leave this as unanswered if you dont mind maybe other people encounter it and help in finding a solution. Should i open an issue instead?

[ryan13mt]

Opened #27