Images · Stories · Voice · Music · Games

Make things on your own graphics card, without the Python hell.

If you have fought ComfyUI, Pinokio, or a conda environment to get one image out, this is the other path. LumaBrowser installs the image engine and a model for you, then adds roleplay, voice, music, and small games on top, all in one free app. Your gaming PC is enough.

Download free What does my card need?
NVIDIA, AMD, and Apple Silicon · no account, no API key · first image in one session

Will it run on my PC?

Chat works on almost anything. Images want a graphics card. The rows below are the same ones Easy Setup uses when it decides what to install.

No graphics card? Images are not recommended. A small model will render on the processor, but a single picture takes a few minutes and you will not enjoy it. Chat, documents, and stories still work fine; add images later if you get a card.

What you can make

Everything here is in the one download. Easy Setup turns images on for you; the rest is a click away inside chat.

Ready

Images and image editing

Type a description, get a picture. Feed it a reference photo to keep the same face or outfit across renders, paint over a region to change it, or upscale. The engine, the model, and the graphics-card driver mode are picked and installed for you.

How image generation works →
Needs a graphics card with 8 GB or more. First model about 6 GB.
Ready

Roleplay and stories

Characters that keep their look from scene to scene, with expressions and outfits that change as the story does. Write the setup or let the AI draft a name, persona, and scenario for you.

Roleplay mode →
Chat model only for text. Add images for character art. Adults only; see the note below.
Ready

Voice conversation

Talk to it and hear the reply, fully offline. Interrupt by talking over it. Nothing recorded leaves the machine.

Voice mode →
Windows and Linux in this release. macOS voice input is not available yet. About 1 GB extra.
Ready

Small games

Describe a game in a sentence ("a top-down dungeon crawler with a torch mechanic") and get something playable, with generated sprites and backdrops. Then ask for changes.

Artifacts and mini-apps →
Chat model for the code. Images on for the art. A mid or large chat model gives much better games.
Preview

Video

Short clips, up to about fifteen seconds, from a description. Slow and hungry; this is the one to try after images are working, not before.

High-end card only (24 GB+). Model downloads in the tens of gigabytes.
Preview

Music

Generate a song from a text description. Present, but not yet exercised widely; treat it as a preview rather than a feature to plan around.

Roughly 34 GB of video memory on one card. Windows needs WSL2; macOS is not supported.

Image models, sizes, and what they need

Easy Setup picks the first one for you based on your card. The others are one click each in the Image tab. Download sizes are what lands on your disk.

ModelGood forDownloadGraphics card
Z-Image TurboA quick everyday default. Eight fast steps, decent text in images, good general quality.about 6 GB8 GB or more
Stable Diffusion 1.5The old classic. Runs on small cards, 512 by 512 native. Lots of community styles.about 2 GB4 GB or more
SDXL 1.0Sharper than SD 1.5, 1024 by 1024 native. A solid all-rounder.about 6.5 GB8 GB or more
FLUX.1 schnellPhotorealistic in four steps. The wizard's high-end pick.12 to 30 GB, depending on variant12 GB or more; 24 GB comfortable
Qwen-Image-EditThe editing model: keep the same person or object across pictures, change one part of an image, follow written edit instructions.about 18 GB24 GB recommended
Anima (add-on)Anime and illustration styles, tag-based prompting.varies8 GB or more

Card numbers are what we recommend for a comfortable experience, not hard limits; the app will try to fit a model on a smaller card by moving parts of it to system memory, at the cost of speed. You can also bring your own .safetensors or .gguf file. Sizes are approximate and depend on the variant chosen. NVIDIA uses CUDA, AMD and Intel use Vulkan, Apple Silicon runs natively; on Linux with NVIDIA, images currently use the Vulkan build.

What to expect in the first session

A gaming PC with a 12 GB card, from download to first image, is about 20 minutes on a normal connection. Most of that is the download.

The download is big and can be paused

A chat model plus an image model is 10 to 50 GB depending on your tier. Downloads resume after a dropped connection or a restart. Nothing starts over.

The first render takes longer

The model loads into the card on first use, which can take a minute. After that a Z-Image Turbo picture on a mid-range card is a few seconds.

Black or blank images

Almost always means the card ran out of memory. Pick a smaller resolution or a smaller model in the Image tab; the image guide lists the fixes.

Antivirus and the installer

The installer is not code-signed yet, so Windows SmartScreen warns once. More info, then Run anyway. No Python, no separate runtime to install, nothing added to your PATH.

Mature content is possible. Roleplay mode and the image models apply no content filter of their own; what they can depict depends on the model you load. These features are intended for adults, and the character pipeline is constrained to adult proportions. You are responsible for the models you choose and the content you generate on your own machine.

First image in one session, then the rest

Download LumaBrowser, click Easy Setup, say yes to images. Then discover roleplay, voice, and games from the same chat box.

Download free Read the image guide