Skip to content

1.6.1
Compare
Choose a tag to compare
@Trojaner Trojaner released this 17 Apr 20:47
· 8 commits to main since this release
1.6.1
b69c265
This commit was signed with the committer’s verified signature.
Trojaner Enes Sadık Özbek
GPG key ID: 85CC9787265BBC7F
Verified
Learn about vigilant mode.

Features

  • Added new tool / function call based generation mode.

This new mode uses tool / function calls to give the LLM model the ability to trigger and generate images itself!
Very similar to how GPT4 calls Dall-E when you request image generation.

Works with no hardcoded triggers, has multilingual support, and uses past chat history context. For example, if you told it earlier to only generate cartoon style images, it will remember this and do it until you say otherwise.

This mode does not support chat streaming and requires special models (like Command-R or Llama-2-chat-hf-function-calls-v2) and some minor system prompts adjustments to describe the tools / function calls available. If you do not use a transformers / HF based loader, you will probably also have to modify start_with so it always outputs JSON. Consult the sample settings.yaml and the model documentation for more information on this.

chrome_1N17C3f54H

image

chrome_mlYtxFsHhO

Bug fixes

  • Added missing "None" option for VAEs
  • Fixed disable streaming during image generation not working.
  • Fixed debug mode not working correctly
  • Replace FaceID checkboxes with dropdown as SD.Next does not support multiple FaceID anymore
  • Removed invalid gradio calls such as refresh_button.update(...)
  • Fixed network requests spam because of wrong gradio usage, this should fix performance issues on the UI side

Misc

  • Improved prompt normalization by removing newlines, hashtag, etc.
  • Replaced mypy with pyright / pylance

DOWNLOAD AND INSTALLATION INSTRUCTIONS HERE

Assets 2
Loading