// the find
camenduru/TostUI
This project is a collection of Docker-based web user interfaces designed to easily run various state-of-the-art generative AI models locally. It simplifies the deployment of these AI tools by packaging them into ready-to-use containers with accessible front-end controls.
A collection of Docker images that wrap generative AI models (image generation, video, 3D, audio) behind a Next.js UI, so you can run them locally with a single `docker run` command. Built by camenduru, who has a long history of packaging AI models for RunPod and local use — the target audience is people who want to try models without setting up Python environments.
Each container is self-contained and starts with one command — no conda, no pip, no CUDA driver wrestling beyond what Docker needs. The 4-bit and 8-bit quantized variants (Z-Image-Turbo 4bit needs only 6GB VRAM) show real consideration for hardware diversity. RunPod serverless templates are included alongside local options, which is a practical choice for people without a 24GB card. The TypeScript frontend is structured as a proper Next.js app with an API route layer rather than a cobbled-together static page.
Almost every model requires 24GB VRAM minimum, which excludes the majority of consumer hardware — only one variant (Z-Image-Turbo 4bit at 6GB) is actually accessible to most people. The repo is a monorepo of UI code but the actual model weights and inference logic live inside opaque Docker images on Docker Hub; you cannot audit what's running or update just the model without pulling a new multi-GB image. No compose files or volume mounts for persisting outputs or custom models — every container run is ephemeral. The README is purely a list of `docker run` commands with no explanation of what each model actually does or how they compare.