Vision Enabled Models

[NotThePassword]

First off, just want to say thank you for a great app! And the other thing I wanted to do was ask if you have thought about the models coming out with vision enabled and possibly added the ability to send images through the chat interface?

[Vali-98]

This has been requested many times, and it really just comes down to two issues:

1. ChatterUI lacks any media storage. We will need to figure out how to attach media to the app cleanly.
2. Vision models are very slow on mobile. Images are sent through an encoder which may not be supported by llama.rn and itself will take up about 200-300 tokens. Given the somewhat slow prefill speeds due to no NPU/GPU utilization, this makes running vision models non-viable, and honestly the results of said models are pretty poor due to their small sizes anyways.

I wont put it as a 'never add'. Models may get smarter and more efficient in the future, but its a 'not useful right now' situation.

[NotThePassword]

Understood. Thank you for taking the time to respond.

[tazztone]

gemma 3n will be possible to run?