Launching today
Logen
Run local AI image models on your GPU, minus the nodes
3 followers
Run local AI image models on your GPU, minus the nodes
3 followers
Desktop app for Windows and Linux x64 that runs AI image models on your own GPU instead of cloud subscriptions or ComfyUI node graphs. Free tier includes FLUX.1-schnell (Apache-2.0). Pro (£49 once) adds FLUX.2 [klein] 4B, Z-Image-Turbo, and batch runs.
Free Options
Launch Team
CloudflareDeploy serverless functions globally in seconds
Promoted
Hi everyone, I'm Molly.
I started building Logen after a weekend where I spent more time fixing a broken PyTorch CUDA install in a Python venv than actually making images. If you own an NVIDIA GPU, you shouldn't have to choose between paying $30/month for cloud credits and staring at a screen full of ComfyUI nodes just to render an asset.
Logen 1.4.0 wraps a managed ComfyUI engine and local Ollama chat into a single desktop app for Windows 10/11 (x64) and Ubuntu/Debian Linux (x64).
When you open it, you work inside chat threads tied to project folders on disk (Documents/Logen/). You type what you want, local Ollama refines the prompt if needed, and ComfyUI renders it on your GPU. Clicking any image opens a full-screen 1:1 pixel lightbox, and you can jump from any saved render straight back to the chat thread that made it. When you want direct control over steps, CFG guidance, seed, or image-to-image denoise without touching nodes, there is a two-column workflow studio where you can save parameter presets.
Version 1.4.0 also fixes the most annoying part of local AI: background services dying silently. If Ollama isn't running when you hit send, Logen starts ollama serve in the background and retries automatically. If a model weight is missing or ComfyUI hits a VRAM limit, the error banner tells you what broke and gives you a one-click button to fix it while keeping your typed prompt intact.
How pricing and model licences work:
The Free tier (£0) runs FLUX.1-schnell Q4 (Apache-2.0 licence) with project chat threads, gallery folders, and the 1:1 lightbox.
The Pro tier (£49 one-time payment, no subscription) adds FLUX.2 [klein] 4B and Z-Image-Turbo INT8 (both Apache-2.0), custom presets, and 2-to-16 seed batch runs, with a 24-hour offline licence validation grace window. We keep non-commercial weight sets (FLUX.1-dev and Qwen-Image-2.1) disabled in the app so you never accidentally mix non-commercial models into commercial work.
One limitation upfront: Logen 1.4.0 is built for 64-bit Windows 10/11 and Ubuntu/Debian Linux with an NVIDIA GPU (6 GB+ VRAM; we test daily on an RTX A2000 12 GB). macOS is not supported.
You can download the free Windows installer or Linux AppImage/DEB at https://titanxt.io/download/. If you try it on your GPU, let me know how the setup wizard and generation times hold up on your card.