Layla v5.4.0 has been published!

This update for Layla adds support to the latest Gemma3 models, LaTeX formulae, and more!

Gemma 3 is the latest open source model by Google: https://blog.google/technology/developers/gemma-3/

It provides multi-modal functionality starting from it's 4B model, and is a perfect size to run on phones!

Layla supports Gemma-3 GGUF models, which you can download here: https://huggingface.co/ggml-org/gemma-3-4b-it-GGUF

LaTeX formula

Layla can now display mathematical formulae during chat.

Image generation improvements

This latest update supports the latest Snapdragon Elite NPU for image generation. Additionally, all image models have been recompiled to give better quality across the board!

Full Changelog

New features:

• Gemma3 family models support!

• NPU image generation now supports the latest Snapdragon Elite CPU

• added new NPU stable diffusion models

• added new CPU stable diffusion models

Improvements:

• Layla supports displaying LaTeX formulae in chat messages

• re-compiled several stable diffusion models with their custom VAE decoders, which will produce better image quality

• backup data will now backup your custom prompts

• turning nsfw option off in settings will also hide nsfw characters in your character selection screen

• added ability to copy generated code in chat messages

• added support for MultiTTS voices

• added ability to continue generating a message from the LLM

Bug fixes:

• fixed a bug where phones without access to Google cannot access Layla Cloud subscriptions

• fixed bug where GPU inference was not working with Background Inference