[bug]: Installing Models when user skipped all during install

[enzyme69]

Is there an existing issue for this?

• I have searched the existing issues

OS

macOS

GPU

mps

VRAM

8 GB

What happened?

On a Mac, during install on a new computer, I skipped all the models and ended up with empty model .yaml. I could not find a way to install model easily. I thought that I could just run the CLI and then import model from there (with no model loaded). Apparently that's not possible and kept getting error, it won't run.

 Current VRAM usage:  0.00G
** "stable-diffusion-1.5" is not a known model name. Please check your models.yaml file
** "stable-diffusion-1.5" is not a known model name. Please check your models.yaml file
╭─────────────────────────── Traceback (most recent call last) ────────────────────────────╮
│ /Users/blendersushi/Documents/InvokeAI/scripts/invoke.py:884 in <module>                 │
│                                                                                          │
│   881 ######################################                                             │
│   882                                                                                    │
│   883 if __name__ == '__main__':                                                         │
│ ❱ 884 │   main()                                                                         │
│   885                                                                                    │
│                                                                                          │
│ /Users/blendersushi/Documents/InvokeAI/scripts/invoke.py:101 in main                     │
│                                                                                          │
│    98 │   │   print(">> changed to seamless tiling mode")                                │
│    99 │                                                                                  │
│   100 │   # preload the model                                                            │
│ ❱ 101 │   gen.load_model()                                                               │
│   102 │                                                                                  │
│   103 │   # web server loops forever                                                     │
│   104 │   if opt.web or opt.gui:                                                         │
│                                                                                          │
│ /Users/blendersushi/Documents/InvokeAI/ldm/generate.py:783 in load_model                 │
│                                                                                          │
│    780 │   │   '''                                                                       │
│    781 │   │   preload model identified in self.model_name                               │
│    782 │   │   '''                                                                       │
│ ❱  783 │   │   self.set_model(self.model_name)                                           │
│    784 │                                                                                 │
│    785 │   def set_model(self,model_name):                                               │
│    786 │   │   """                                                                       │
│                                                                                          │
│ /Users/blendersushi/Documents/InvokeAI/ldm/generate.py:808 in set_model                  │
│                                                                                          │
│    805 │   │   if model_data is None:  # restore previous                                │
│    806 │   │   │   model_data = cache.get_model(self.model_name)                         │
│    807 │   │                                                                             │
│ ❱  808 │   │   self.model = model_data['model']                                          │
│    809 │   │   self.width = model_data['width']                                          │
│    810 │   │   self.height= model_data['height']                                         │
│    811 │   │   self.model_hash = model_data['hash']    

Now, trying to install the model manually...

scripts/preload_models.py

**Error creating config file ./configs/models.yaml: [Errno 2] No such file or directory: './configs/new_config.tmp' **

Screenshots

No response

Additional context

No response

Contact Details

No response

[enzyme69]

The only to fix this "skipped model install" is to manually type the path into config/model.yaml

Would be nice instead of erroring when running: it will tell user: hey you don't have any stable diffusion model ckpt, would you like to import one and set it up for you

redshift:
  weights: /Users/blendersushi/Downloads/_StableDiffusion_CKPT/redshift-diffusion-v1.ckpt
  description: redshift render like cinema4D style
  config: configs/stable-diffusion/v1-inference.yaml
  width: 512
  height: 512
  default: false
cats_musical:
  weights: /Users/blendersushi/Downloads/_StableDiffusion_CKPT/Cats-Musical-Style-ctsmscl.ckpt
  description: 'cats musical '
  config: configs/stable-diffusion/v1-inference.yaml
  width: 512
  height: 512
  default: false

[lstein]

I retracted my previous. This is a bug in preload_models.py. It shouldn't crash out on you.

[lstein]

I can't reproduce the problem of preload_models.py not being runnable. Is there any chance that your configs directory cannot be written?

I'll see if I can fix invoke.py to let you install a model if there isn't one already configured.

[lstein]

I've just created a pull request that will fix this issue. If no config.yaml file is present, invoke.py will offer to create one for you using a provided models file. It's a little rough and dirty (did it in a rush), but better than the current crash.