Layla v6.11.0 has been published!
- Layla

- May 20
- 4 min read
One-Shot Voice Cloning, AI Agents in Companion Mode, and Full-Screen Chat
Layla — the offline AI assistant — now supports PocketTTS!
You can now clone any voice from a single audio sample, run AI agents from your phone's default assistant or directly inside companion mode, and chat with your companion in full-screen.
Clone any voice in seconds with PocketTTS
Layla now supports one-shot voice cloning through the new PocketTTS mini-app. "One-shot" means exactly what it sounds like: give Layla a single short audio sample and it produces a custom voice you can assign to any character, set as the global TTS voice for the entire app, or configure per-character.

You can build a voice two ways:
Record directly in Layla — open PocketTTS, hit record, and capture a sample without leaving the app
Upload a file — bring your own audio sample from anywhere on your phone

Once the voice exists, it's yours to use offline forever. No cloud, no API costs, no monthly subscription.
Tips for the best voice cloning results
PocketTTS is fast but it learns from exactly what you give it, so sample quality matters more than length:
Record in a quiet room. PocketTTS is very sensitive to background noise — if you can't avoid it, run your clip through an online noise-removal tool before importing.
Use 16-bit PCM WAV files only. MP3 and other formats won't work — convert them with any online audio converter first.
Avoid silences in your sample. PocketTTS will learn the gaps too and reproduce them as long, awkward pauses.
Exaggerated voices clone better. Distinctive, slightly theatrical samples produce more recognizable cloned voices than flat, monotone ones.
Pro tip: get studio-quality voice clones for free with ElevenLabs
If you want clean voice clones for Layla without the AI sounding like it's having an asthma attack, exploit the ElevenLabs free tier. You get 10,000 characters per month, and the Voice Library has 10,000+ community voices you can use as a base — for zero dollars. Here's the workflow:
Make a free ElevenLabs account and pick a base voice from the Voice Library.
Write a script of around 145 words. Zero punctuation. No commas, no periods, no nothing. If you add punctuation, the AI injects dead air — and PocketTTS will faithfully clone that silence as a stutter.
Generate it. The lack of punctuation forces ElevenLabs to spit out a continuous 20–40 second audio block with no pauses.
Download the MP3 and run it through any free online converter to turn it into a 16-bit PCM .wav file.
Drop the .wav into Layla's PocketTTS mini-app. Done.
You now have a flawless studio-quality voice clone running fully offline on your Android phone, for zero dollars.
AI agents are now everywhere in Layla
Layla's Agents — small task-running scripts that can chain tools, search files, run code, and more — used to live inside the main chat. This update puts them everywhere you actually use Layla on your phone:
As your phone's default assistant — set Layla as Android's default assistant and you can now trigger agents from anywhere on your phone, including with a long-press of the home button or via your phone's assist gesture. This makes Layla a true on-device AI personal assistant with no subscription, no cloud, no data leaving your phone.
Inside companion mode — your companion can now run agents too, so the floating companion icon becomes a genuinely capable assistant rather than just a chat window.
Full-screen chat for companion mode
Companion mode previously gave you a small floating chat bubble. You can now expand it into a full-screen view that mirrors the main Layla chat experience — same features, same layout, just launched from your companion.
Three ways to open it:
Double-tap the companion floating icon
Long-press the companion floating icon
Tap the single message bubble
Customise your chat bubble colours
A small but much-requested addition: you can now configure chat-bubble colours directly in UI settings. Match Layla to your phone's wallpaper, your character's theme, or whatever you like.
Improvements
Generated image gallery is now backed up
Flipped the user/AI position of chat bubbles to match the standard convention used by other social media and chat apps
Image gallery now blocks other apps from screenshotting or recording its contents, and automatically hides its contents when Layla goes into the background
Image gallery supports multi-selection for deleting images
Improved TTS latency by implementing audio streaming
You can re-order characters inside folders and drag them out to remove them from the folder
Improved character impression generation by limiting max output and cutting off incomplete sentences
Stable Diffusion prompts are now saved with generated images in the gallery
Bug fixes
Fixed Layla Cloud image generation failing to read generated images after a few attempts
Fixed several crashes when switching between Live2D characters quickly
Fixed default Automatic1111 JSON body no longer working
Fixed max response length not trimming incomplete sentences
Fixed native TTS preview (Google / iOS) not using the selected voice in other languages
Fixed asterisks not being skipped in TTS when the skip-asterisk setting was on
Fixed inability to use a generated image as the input for the next img2img generation
As always, every feature above runs fully offline on your Android device — your voices, characters, conversations, and generated images stay on your phone.



Comments