DEV Community

David
David

Posted on

Locally Uncensored v1.5.5 — Zero-Config AI Models, VRAM Detection & Production-Ready Desktop App

We just shipped v1.5.5 of Locally Uncensored — the open-source, local-first AI desktop app for Chat + Image Generation + Video Generation. No cloud, no censorship, no tracking.

Here's what's new.

Zero-Config Model Experience

The biggest pain point with local AI? Configuration. Which sampler, how many steps, what CFG scale?

Gone. LU now auto-detects your model type and applies optimal defaults. Select a model, hit generate. That's it.

On top of that, a pre-flight validation checks if you have the right VAE, CLIP encoder, and custom nodes before generation starts. Missing something? Direct download buttons right in the error message.

VRAM-Based Recommendations

LU reads your GPU VRAM via ComfyUI and tags every model bundle:

  • Fits your GPU — runs comfortably
  • ⚠️ Needs more VRAM — might struggle

Bundles are sorted by fit. No more guessing if your 8GB card can handle FLUX.

2026 State-of-the-Art Models

Updated model bundles with the latest:

  • FLUX 2 Klein 4B (8 GB) — fast image gen on consumer GPUs
  • LTX Video 2.3 22B FP8 (35 GB) — high-quality video generation
  • Curated text models: GLM 4.6, Qwen 3 front and center

Every bundle includes Model + VAE + Text Encoders — complete packages, one-click download.

Download Manager

Pause, cancel, resume. HTTP Range headers for proper resumable downloads. No more restarting 20GB downloads because your connection dropped.

RAG Document Chat — Fixed

v1.5.0 introduced RAG but it had issues. All fixed:

  • IndexedDB persistence — embeddings survive page reload
  • React 19 compat — no more infinite loops
  • ErrorBoundary — errors show a message, not a white page

Desktop App — Production Ready

The Tauri v2 standalone .exe got a major overhaul:

  • Splash screen with logo while React loads
  • CORS proxy through Rust (no more blocked requests)
  • Whisper loads in background (fast startup)
  • Terminal windows hidden, DevTools only in debug

Bug Fixes

10+ fixes including CLIP/VAE model fallback, BM25 search scoring, agent image generation actually calling ComfyUI, Whisper detection, and chat history filtering.


GitHub — star if you find it useful

📋 Full Changelog

🌐 Landing Page

Stack: React 19 + TypeScript + Tailwind CSS 4 + Vite 8 + Tauri v2 (Rust)

Top comments (0)