Add new console frontend to initial model selection, and other model mgmt improvements

[lstein]

Major Changes

The invokeai-configure script has now been refactored. The work of selecting and downloading initial models at install time is now done by a script named invokeai-model-install (module name is ldm.invoke.config.model_install)

Screen 1 - adjust startup options:

Screen 2 - select SD models:

The calling arguments for invokeai-configure have not changed, so nothing should break. After initializing the root directory, the script calls invokeai-model-install to let the user select the starting models to install.

invokeai-model-install puts up a console GUI with checkboxes to indicate which models to install. It respects the --default_onlyand--yes` arguments so that CI will continue to work. Here are the various effects you can achieve:

invokeai-configure
This will use console-based UI to initialize invokeai.init,
download support models, and choose and download SD models

invokeai-configure --yes
Without activating the GUI, populate invokeai.init with default values,
download support models and download the "recommended" SD models

invokeai-configure --default_only
Activate the GUI for changing init options, but don't show the SD download
form, and automatically download the default SD model (currently SD-1.5)

invokeai-model-install
Select and install models. This can be used to download arbitrary
models from the Internet, install HuggingFace models using their repo_id,
or watch a directory for models to load at startup time

invokeai-model-install --yes
Import the recommended SD models without a GUI

invokeai-model-install --default_only
As above, but only import the default model

Flexible Model Imports

The console GUI allows the user to import arbitrary models into InvokeAI using:

1. A HuggingFace Repo_id
2. A URL (http/https/ftp) that points to a checkpoint or safetensors file
3. A local path on disk pointing to a checkpoint/safetensors file or diffusers directory
4. A directory to be scanned for all checkpoint/safetensors files to be imported

The UI allows the user to specify multiple models to bulk import. The user can specify whether to import the ckpt/safetensors as-is, or convert to diffusers. The user can also designate a directory to be scanned at startup time for checkpoint/safetensors files.

Backend Changes

To support the model selection GUI PR introduces a new method in ldm.invoke.model_manager called `heuristic_import(). This accepts a string-like object which can be a repo_id, URL, local path or directory. It will figure out what the object is and import it. It interrogates the contents of checkpoint and safetensors files to determine what type of SD model they are -- v1.x, v2.x or v1.x inpainting.

Installer

I am attaching a zip file of the installer if you would like to try the process from end to end.
InvokeAI-installer-v2.3.0.zip

[lstein]

I've decided to turn this into a full-featured interface for importing models from the Internet. I'm turning it into a draft while working out the bugs. Currently it is nonfunctional.

[mauwii]

LGTM!

[psychedelicious]

Sorry for the late review. I had tried to review a few days ago but I couldn't get it to run; I think you were in the middle of changes.

First impression - a thing of beauty!

Running the script, I had a few hiccups:

• Before running the script, I had dreamlike-diffusion-1.0 and dreamlike-photoreal-2.0 downloaded but not in my models.yaml. I had ft-mse-improved-autoencoder-840000 downloaded and present in my models.yaml.

None of these were picked up as installed - I suppose the script looks at models.yaml to see what is installed.

After selecting dreamlike-diffusion-1.0 and autoencoder-840000, the script detected they were downloaded correctly and added them to my models.yaml. However, I now have two nearly identical entries for the VAE: ft-mse-improved-autoencoder-840000 and now autoencoder-840000, pointing to the same ckpt. Only difference is the name.

I did not select dreamlike-photoreal-2.0 to be installed, and it did not show up in my models.yaml after the script. I would expect the script to pick up everything previously installed by InvokeAI.

• At the end of the first run, my invokeai.init file had not changed. After the second and subsequent runs, it had changed to match my selections...
• ...almost - no matter what I select for convert to diffusers, nothing changes in invokeai.init
• Finally, the setting the NSFW box enabled or disabled does change invokeai.init, but it always defaults to enabled when running the script, even if it is disabled in invokeai.init

This is a very nice user experience for a shell script, great work! Love it.

[lstein]

trying this out now. i expect the == CONVERT IMPORTED MODELS INTO DIFFUSERS == toggle will confuse people - is it possible for it to simply not appear if i'm not importing anything?

also, can the == IMPORT LOCAL AND REMOTE MODELS == text entry box be smaller? it really only needs to be 1 line i think..

Changes implemented:

1. The convert toggle is now conditional and only appears when there are models to import or when the "autoscan" directory is set.
2. I have made the text entry box smaller. However, I want to make it possible for people to download multiple models at a time, and so I made it 5 lines high rather than 1 line as suggested

[lstein]

Sorry for the late review. I had tried to review a few days ago but I couldn't get it to run; I think you were in the middle of changes.

First impression - a thing of beauty!

Running the script, I had a few hiccups:

• Before running the script, I had dreamlike-diffusion-1.0 and dreamlike-photoreal-2.0 downloaded but not in my models.yaml. I had ft-mse-improved-autoencoder-840000 downloaded and present in my models.yaml.

None of these were picked up as installed - I suppose the script looks at models.yaml to see what is installed.

The script only works off of what's in models.yaml because in fact you can have models installed anywhere in your filesystem. I'll fix it in the future so that the invokeai/models directory is scanned if that's the desired behavior.

After selecting dreamlike-diffusion-1.0 and autoencoder-840000, the script detected they were downloaded correctly and added them to my models.yaml. However, I now have two nearly identical entries for the VAE: ft-mse-improved-autoencoder-840000 and now autoencoder-840000, pointing to the same ckpt. Only difference is the name.

The autoencoder is no longer an option to install. Both the legacy and the diffusers versions are now installed behind the scenes for use in later model importation.

I did not select dreamlike-photoreal-2.0 to be installed, and it did not show up in my models.yaml after the script. I would expect the script to pick up everything previously installed by InvokeAI.

See above. The question is what to do when the user has a models directory that contains model files that are not in models.yaml. Do they want to import these back into models.yaml, or was there a reason they removed them in the first place.

• At the end of the first run, my invokeai.init file had not changed. After the second and subsequent runs, it had changed to match my selections...
• ...almost - no matter what I select for convert to diffusers, nothing changes in invokeai.init
• Finally, the setting the NSFW box enabled or disabled does change invokeai.init, but it always defaults to enabled when running the script, even if it is disabled in invokeai.init

You found a bug. Fixed.

This is a very nice user experience for a shell script, great work! Love it.

Very high praise coming from you! Thanks.

[lstein]

oh, i see i'm supposed to run invokeai-configure rather than invokeai-model-download. hmm.

1. on a fresh vast.ai box if i run invokeai-configure, deselect all the default models and only enter a huggingface repo id in the text box, i end up with an empty models.yaml.
2. even with everything deselected, a lot of stuff gets downloaded, including a 1.7GB CLIP model that i don't need because it comes packaged with the diffusers repo id, and a bunch of face gen stuff that i have never used and likely will never use. is there any way to be smarter about this and/or make it an option?

I have fixed the issue of empty models.yaml files when no recommended starter files are selected. Thanks for detecting it.

[lstein]

@psychedelicious @damian0815 Thank you for your reviews. I have addressed your concerns and would welcome another round of reviews.

[lstein]

above was fixable with mkdir -p /root/invokeai/configs && touch /root/invokeai/configs/models.yaml.

This should no longer be necessary. If the model installer finds that there is no root configs or models directory, it will launch the invokeai-configure command in order to set up the root directory properly. The model installer will then pick up where it left off.

i deselected all the default models, entered a HF repo id in the text entry box and hit DONE and it exited immediately. and /root/invokeai/configs/models.yaml is empty. that's not good.

This bug is fixed.

[damian0815]

thanks!

[psychedelicious]

• Resizing works, but did crash on me. I ran the script having enlarged my terminal window. Then shrunk it down a bit and it crashed due to not enough window space. If disabling the responsive will prevent crashes, I think that's better than introducing a failure condition. If resizing too small was already a failure condition then responsive is fine.
• NSFW option is detected ✔️
• Unfortunately there is a new bug, appears to be related to the responsive feature. Often (but not always), the last few settings never appear and I just get a blank UI. I didn't catch it on the video, but in this state, when resizing my terminal, the missing settings briefly flash on the screen as the UI resizes. Furthermore, when this is occurring, resizing can cause a different error:

A problem occurred during initialization.
The error was: "list index out of range"
Traceback (most recent call last):
  File "/home/bat/Documents/Code/InvokeAI/ldm/invoke/config/invokeai_configure.py", line 828, in main
    init_options, models_to_download = run_console_ui(opt, init_file)
  File "/home/bat/Documents/Code/InvokeAI/ldm/invoke/config/invokeai_configure.py", line 683, in run_console_ui
    editApp.run()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/apNPSApplication.py", line 30, in run
    return npyssafewrapper.wrapper(self.__remove_argument_call_main)
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/npyssafewrapper.py", line 41, in wrapper
    wrapper_no_fork(call_function)
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/npyssafewrapper.py", line 97, in wrapper_no_fork
    return_code = call_function(_SCREEN)
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/apNPSApplication.py", line 25, in __remove_argument_call_main
    return self.main()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/apNPSApplicationManaged.py", line 172, in main
    self._THISFORM.edit()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/fm_form_edit_loop.py", line 47, in edit
    self.edit_loop()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/fm_form_edit_loop.py", line 38, in edit_loop
    self._widgets__[self.editw].edit()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgwidget.py", line 458, in edit
    self._edit_loop()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgwidget.py", line 474, in _edit_loop
    self.get_and_use_key_press()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgwidget.py", line 610, in get_and_use_key_press
    self.handle_input(ch)
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgwidget.py", line 95, in handle_input
    if self.parent.handle_input(_input):
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgwidget.py", line 71, in handle_input
    self.handlers[_input](_input)
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/fmFormMultiPage.py", line 34, in _resize
    w._resize()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgwidget.py", line 326, in _resize
    self.resize()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgtitlefield.py", line 82, in resize
    self.entry_widget._resize()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgwidget.py", line 326, in _resize
    self.resize()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgmultiline.py", line 105, in resize
    self.display()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgwidget.py", line 429, in display
    self.update()
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgselectone.py", line 18, in update
    super(SelectOne, self).update(clear=clear)
  File "/home/bat/invokeai/.venv/lib/python3.10/site-packages/npyscreen/wgmultiline.py", line 220, in update
    line = self._my_widgets[-1]
IndexError: list index out of range

Video of missing settings:
Screencast from 22-02-23 18:55:21.webm

If the cause isn't clear, maybe we should skip the resizing for this release, in which case I am happy to approve this.

[psychedelicious]

Also, somehow while testing this I ended up with a wonky invokeai.init file:

# InvokeAI initialization file
# This is the InvokeAI initialization file, which contains command-line default values.
# Feel free to edit. If anything goes wrong, you can re-initialize this file by deleting
# or renaming it and then running invokeai-configure again.
# the --outdir option controls the default location of image files.
# generation arguments
# You may place other  frequently-used startup commands here, one or more per line.
# Examples:
# --web --host=0.0.0.0
# --steps=20
# -Ak_euler_a -C10.0
#
--no-ckpt_convert
--no-ckpt_convert
--no-ckpt_convert
--no-ckpt_convert
--no-ckpt_convert
--no-ckpt_convert

--outdir=/home/bat/invokeai/outputs
--embedding_path=/home/bat/invokeai/embeddings
--precision=auto
--max_loaded_models=2
--no-nsfw_checker
--xformers
--no-ckpt_convert

[psychedelicious]

👍