AI-Powered
Media Studio
Background removal, upscaling, stem separation, and image generation — all running locally on private iMac hardware via Tailscale.
Running on local Intel iMac GPU. Process is private and secure. Do not close browser.
Prompt Guide & Settings ▾
AI models are naturally "stubborn" in img2img. You need a highly specific positive prompt, a ruthless negative prompt, and the right denoising strength.
Prompt Tip
Use detailed descriptions. Set denoising to 0.55–0.70 for best results. CFG Scale 7.5–9 for strict prompt adherence.
The Self-Hosted
AI
Media Server Revolution
Four powerful AI models running directly on private hardware — no cloud subscriptions, no data leaving your network. Background removal via InSPyReNet, upscaling through RealESRGAN, audio stem separation with Meta Demucs, and image generation via Stable Diffusion, all orchestrated through a single Tailscale-secured endpoint.
Processing Topology
Each model runs locally on the same Intel iMac with an AMD Radeon Pro 5300 GPU. The server is exposed via Tailscale Funnel (HTTPS), ensuring all processing happens on your own hardware with no third-party inference.
Why Self-Host?
While cloud AI platforms offer convenience, they come at the cost of privacy, recurring subscription fees, and usage limits. This server runs entirely on dedicated hardware — an Intel-based iMac with 40GB of RAM and a dedicated AMD GPU. Background removal, upscaling, Demucs stem separation, and Stable Diffusion inference all execute locally, accessed securely through Tailscale Funnel from any browser worldwide. No data leaves this machine. Check the README.md for steps on how to clone this repo and host it on your own hardware.
The Stack: Python 3.10 + PyTorch 2.2.2 running on macOS. The Gradio-based server orchestrates all four model pipelines behind a unified interface. Tailscale Funnel provides TLS termination and global routing, while Firebase manages authentication and usage statistics.