Add text prompt to inpaint mask support

[lstein]

add clipseg support for creating inpaint masks from text

On the command line, the new option is --text_mask or -tm.
Example:

invoke> a baseball -I /path/to/still_life.png -tm orange

This will find the orange fruit in the still life painting and replace it with an image of a baseball.

[tildebyte]

Is 528a183 still necessary?

[lstein]

Are there any issues you'd like to see resolved before I merge this?

[lstein]

I'm going to go ahead and merge before too many new conflicts arise.

[lstein]

The preload script is just loading each of the models in turn and then discarding them. This seems senseless until you realize that these are Hugging Face models that are loaded on demand via URL. If you are behind a firewall, like me, then Invoke will fail whenever it tries to load one of these models. So we use the preload script that downloads the models as a side effect and caches them in local directories. Lincoln

…

On Tue, Oct 18, 2022 at 1:12 PM tildebyte ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In scripts/preload_models.py <#1133 (comment)>: > transformers.logging.set_verbosity_error() # this will preload the Bert tokenizer fles -print('preloading bert tokenizer...', end='') - -tokenizer = BertTokenizerFast.from_pretrained('bert-base-uncased') +print('Loading bert tokenizer (ignore deprecation errors)...', end='') +with warnings.catch_warnings(): + warnings.filterwarnings('ignore', category=DeprecationWarning) + tokenizer = BertTokenizerFast.from_pretrained('bert-base-uncased') Right, but the BERT version of the tokenizer is permanently overwritten by the CLIP version, so *nothing* is "using" it currently. Relevant snippet (note that there's no conditional, nor is BERT "handed off" to another alias of some kind): # this will preload the Bert tokenizer flesprint('preloading bert tokenizer...', end='') tokenizer = BertTokenizerFast.from_pretrained('bert-base-uncased')print('...success') ...*snip 7 irrelevant lines* ... print('preloading CLIP model...',end='')sys.stdout.flush() tokenizer = CLIPTokenizer.from_pretrained(version) — Reply to this email directly, view it on GitHub <#1133 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAA3EVIY4DHESH37MR4UO23WD3K7HANCNFSM6AAAAAARGU4YUI> . You are receiving this because you modified the open/close state.Message ID: ***@***.***>

[Any-Winter-4079]

This is an amazing new feature.
Works surprisingly well with Textual Inversion embeddings.