GitLab

fix model v2 link,default is v1 link

* fix checkpoint path

* fix checkpoint path in inference_pmv2_controlnet

* Fix typo in comment

* Enable choosing output aspect-ratio

* Explain the effect of aspect ratio in the README

---------
Co-authored-by: Markku Post <markku.post@cleverest.eu>

* add capability to interrupt the pipeline

* Update pipeline.py

---------
Co-authored-by: Zhen Li <zhenli1031@gmail.com>

* Add working cog demo

* Add safety checker

* Print seed when doing a prediction

* Update README.md

* Update for upstream changes, update torch==2.1.1, pget model weights

* Add up to four inputs and outputs in cog demo

And clean up predict.py based on review feedback

* Add some debug output for users

* Improve cog parameters based on feedback

* Strip trigger_word from prompt using tokenizer

* Fix guidance scale bug, allow no classifier free guidance in pipeline.

* Update README.md

* Disallow trigger word in negative prompt

---------
Co-authored-by: Judith van Stegeren <judith@replicate.com>
Co-authored-by: Judith van Stegeren <690008+jd7h@users.noreply.github.com>

The modification replaces this method with a token-based approach using a tokenizer. This tokenizes the entire prompt, identifies the token corresponding to the trigger word, removes it, and then decodes the modified tokens back into text.