Locally Uncensored, the plug-and-play AI studio·Chat · Code · Image · Video · Remote
An uncensored AI answers the question instead of refusing it. You can run one free on your own PC, or open it in a browser when the GPU is not there. Read what uncensored AI means here.
v3.0.3 · WINDOWS · ~14 MB installer · 90,000+ DOWNLOADS
Free and open source. On Linux, take the AppImage or a package from GitHub. No GPU, or on a Mac? Use LU Cloud.
What you can do
Real AI on your own computer. One installer, no command line, no Docker. Private, free, and yours. In local mode there is no account, no cloud and nobody watching.
Ask anything and you get an answer instead of a lecture. Show it a photo, drop in a PDF, and it remembers what matters across chats. Ask for a picture and it draws one right in the conversation. The app finds and installs the right free model for your PC, so all you do is type.
Type what you imagine and get a picture. Change it with words, cut it out, or turn it into a clip. Twelve modes share one tab, and everything you make lands in your own gallery. The app only offers what your computer can handle and installs each tool with one click.
All twelve run on LU Cloud. Enhance and Erase need the cloud; the other ten also run on your own graphics card.
Point the Coding Agent at your project and describe the problem in plain words. It reads the files, says what it found, and shows every change as a diff before anything touches your code. Here it catches a temperature formula that is off by 0.15. It even runs on small models most PCs can handle.
Give Agent Mode a goal and it picks its own tools. It searches the web, writes the files and runs each step while you watch. Ask for a website and you get one, built in a sandbox on your own machine.
Your PC does the heavy lifting and your phone is the remote. Scan a QR code, type a 6-digit code, and chat, coding and tools are in your pocket, at home or on the road. Remote access stays off until you switch it on.
No GPU, or the wrong one
The hosted side opens in a browser, which is also how it works on a Mac and on a phone, and it is paid from credit packs that start at 15 EUR, never expire and need no subscription.
We add no moderation layer of our own to text. For cloud image and video, your LU Cloud account has a content policy setting: Strict, Standard, or off.
47 chat models, every one of them included in every plan and every pack. We add no moderation layer of our own to text.
Chat in the browserAdult-capable image models cost from 1,000 credits per picture, and the account has to switch its own content policy setting first.
See the image modelsHosted video is image-to-video only: you supply the starting frame, and each clip is five seconds long. The input image is not included.
See the video modelsAn OpenAI-compatible endpoint at https://lu-labs.ai/api/inference/v1. Keys begin with lu_, up to five stay active, and every call spends credits.
Read the API pageNew in 3.0.1
3.0.1 is a correctness release with two additions. Qwen-Image 2.1 makes a picture from a prompt and edits one from a reference image, on your own card, on Windows and Linux. LoRAs get a tab of their own in Models. On the fixing side: a graphics card whose free memory could not be measured is planned more carefully instead of optimistically, the Linux AppImage stops handing its own runtime to every program LU starts, and the local engine answers one conversation at a time and says how many are waiting. Free as always, and the app updates itself.
Write a prompt and get a picture, or add a reference image plus a prompt and edit one, with no mask to paint. It appears in the Generate picker and the Edit picker like any other installed image model. The bundle pulls about 16 GB across three files and needs ComfyUI 0.37.0 or newer; an older ComfyUI is told which version to update to first. One reference image at a time, and no painted masks, because the encode node it runs on takes neither.
Get searches CivitAI and downloads what you pick straight into ComfyUI's loras folder. Installed lists what is on disk with its size, marks a character you trained yourself with its trigger word, and deletes a file you no longer want. Before this they sat unnamed among the checkpoints in the Image tab. In Create, a character LoRA now offers only the models trained for that character's family, so the picker cannot suggest a pairing that would fail.
When no free VRAM reading was available, a graphics card used to be planned as if the whole card were sitting empty, and the log said "N MiB are free" for a number that was really the total size, other programs included. That weaker reading now gets a bigger safety margin and is logged as total capacity, so a start places fewer layers rather than too many. The engine also picks its processor code path at startup, so a machine without AVX2 runs local models instead of exiting straight after start, and says which instruction set is missing when it cannot.
git, a system Python, pip, ffmpeg, nvidia-smi, the coding agent shell and everything the Character Trainer starts no longer inherit the AppImage runtime's own library and Python variables. That inheritance made a perfectly healthy system Python fail to find its own standard library, and the diagnosis pointed at the wrong thing. On a distribution where pip refuses to write into the system Python, the isolated environment LU already built survives the first pip call, and the Coding Agent's terminal no longer picks up the same broken environment.
The local model answers one chat at a time. A chat that has to wait now says so, with a line showing how many are ahead of it, and Stop works while it waits and takes it out of the line. Stop, signing out and quitting reach every conversation, not only the one on screen. Temperature, Top P and Max tokens belong to the conversation you moved them in instead of every open chat sharing one value from Settings, and a field nobody moved is left out of the request so the model applies its own default.
A cloud video render now offers exactly the lengths the chosen model supports, read from the live catalog, instead of the same fixed pair for every model. A render started without a prompt, a presenter reading a script for instance, gets a real name in the gallery instead of an empty tile. The catalogue itself is unchanged in this release, and so is the content policy setting in your LU Cloud account: Strict, Standard, or off. It applies to cloud image and video.
Setup
No tech skills needed. There is no command line and there are no setup files. Install it like any normal app and you're chatting in minutes.
One file. Click it and install like any app. It keeps itself up to date. Free, forever.
On first start it checks what you've already got and sets it up for you. Nothing installed yet? It grabs the free AI engine for you in one click.
Pick an AI and start typing. Flip to coding, or to images and video, whenever you like. Want to add a cloud account later? One setting.
The full handbook: install, first start, chat, agent, Create, Cloud →
Models
Whatever you want to do, whether that is chat, art or video, there's a free AI for it. The app installs them for you and only suggests the ones your computer can run. The big names are all here: Qwen, GPT-OSS, Llama 4, Gemma 4, DeepSeek and more.
Google's free chat AI. It can read your pictures too, and the small version runs on a basic PC.
The smart ones, best for thinking through tough questions and writing code. Uncensored versions are included that just answer, with no lectures.
A crowd favourite for photo-real images. Make a picture from words, or remix one you already have.
Truly uncensored, no filters, and about 10 seconds per image.
Turn a sentence into a video clip. Quick, and works on regular hardware.
Got a picture? Turn it into a moving video. Runs on most laptops with a graphics card.
Why
In local mode your chats never leave your computer. No subscription, no tracking, no limits, and no company reading along. Most local AI apps stop at text chat; this one gives you chat, code, images, and video in one place, and your phone can run it from anywhere.
It's free and open source, so anyone can check the code. Local wins for privacy and unlimited use; when you want a cloud model for the really hard stuff, you can plug your own in. Tested with 2,200+ automated checks so it just works.
Writing
Plain-English guides to running AI on your own PC: what to install, what runs on your machine, and how it compares to the rest.
No command line, no Docker. One installer, one-click models, honest hardware requirements.
ComparisonLocally Uncensored, Jan, Ollama, Open WebUI, GPT4All, Msty and KoboldCpp, compared honestly.
BeginnerFLUX and SDXL quality on your GPU without touching a ComfyUI node graph.
GuideYour PC runs the model, your phone is the remote. Private, free, no APK needed.
GuideReal GGUF sizes, the VRAM math for 12 GB and 24 GB cards, the chat template trap, vision setup.
GuideQwen 3.6, Gemma 4 Heretic, Llama 3.1, Hermes 3. What abliteration is, where to download.
ReleaseSingle-instance lock, configurable HuggingFace path, in-app Privacy section, Linux drag fix.
GuideAll sizes from E4B to 27B. Native tools, vision, uncensored variants.
GuideUpload a photo and turn it into something new. Powered by FLUX, Z-Image, SDXL.
ComparisonComplete comparison of GPT4All, Open WebUI, LM Studio, Jan, and more.
GuideSetup guide. Models, hardware, and why local beats cloud.
VersusBoth open source. Only one does chat + code + images + video.
VersusOpen source all-in-one vs polished closed-source chat client.
GuideHow LU handles ComfyUI setup, models, and workflows automatically.
GuideWan 2.1, HunyuanVideo, LTX, FramePack. Hardware requirements, model picks.
Questions
A free, open-source local AI studio for your desktop. Install it like any normal app and you're chatting, generating images, and making videos in minutes, with no command line and no Docker. Under the hood: chat with 20+ provider presets, a coding agent, image generation via ComfyUI (FLUX 2, Juggernaut XL, Z-Image, SDXL), and video generation (Wan 2.1, HunyuanVideo, LTX 2.3, FramePack F1) in one interface. AGPL-3.0 licensed.
Yes. Qwen 3.8 27B is in the Model Manager since v2.6.6, and vision works: the image tower comes as a separate mmproj file that the download writes next to the model. Official, abliterated and uncensored builds, plus a 9B distill for small cards. GPT-OSS-120B and GPT-OSS-20B run via Ollama. GLM-4.7 Flash supported through Ollama. Also ready: DeepSeek R1, Llama 4, Gemma 4, Mistral Small 3, Phi 4.
Yes. Locally Uncensored works as a ChatGPT and Claude alternative that runs on your own hardware. Use Qwen 3.8, GPT-OSS, GLM-4.7, DeepSeek R1, Llama 4 or Gemma 4 instead, or add cloud providers (OpenAI, Anthropic, OpenRouter, Groq) alongside the local stack.
Yes. After setup and model download, no internet is needed for the local providers. No accounts, no telemetry, no usage limits. Cloud providers are optional, and the core runs one-hundred percent on your hardware.
Those tools handle text chat. Locally Uncensored adds a coding agent with fourteen MCP tools, image generation, video creation, A/B model comparison, local benchmarking, granular permissions, file upload with vision, and thinking mode, all in one app.
Text chat: 8 GB RAM. Image generation: an NVIDIA card from 6 GB of VRAM with the starter checkpoint, 8 to 10 GB for FLUX. Video generation: from 6 GB for AnimateDiff and image to video with FramePack F1, from 8 GB for the Wan 2.1 starter, 12 GB and up for Wan 2.2 and HunyuanVideo. The app auto-detects hardware and recommends models. Windows 10/11 and Linux supported.
Abliterated models with artificial restrictions removed. The AI responds honestly without refusing or adding disclaimers. Combined with local execution, your conversations stay private.
Only if you explicitly dispatch a chat over LAN or Cloudflare Tunnel. Remote is opt-in, gated behind a six-digit passcode, and you see exactly when a device is connected. No background uploads. No telemetry.
No. There is no Mac build and there has never been one. The release workflow builds Windows and Linux. On a Mac, use the hosted studio in a browser at lu-labs.ai.
Chat, create and code. On your PC or in the cloud.
v3.0.3 · WINDOWS · ~14 MB installer · 90,000+ DOWNLOADS · 100% open source
Hosted includes 11 Flash models; Pro and Max include 12. Up to 500,000 combined input and output tokens per account per UTC day in app sessions. GLM 5.3 Flash uses credits on Hosted and joins the daily allowance on Pro and Max. API keys always use credits.
EUR 19/month, 900,000 shared monthly credits: up to 900 Neta Lumina Spicy images or 90 five-second LTX 2.3 Spicy clips. Every output spends the same wallet.
Each maximum spends the entire shared monthly wallet on one model. These are alternatives, not separate allowances. Input-image generation, character training, LoRA routes and additional operations cost extra. Uncensored video uses an input image and five-second clips. Failed jobs and changing catalog prices can affect results.
Image models and exact quantities · Video models and exact quantities · Venice comparison · Current facts JSON