I’ve been burning money per image on a rented server, and every time I look at the GPU sitting idle in my own rig, I feel like a total sucker. So I tinkered with this open-source Off Grid AI Desktop and got DreamShaper XL Turbo running locally—works on both Mac and PC, and you can generate images offline too. The Turbo version is distilled, so while regular SDXL needs like 20-30 steps, this one spits out a decent image in just a few steps. The speed is noticeably faster. The model is a quantized GGUF package, so it doesn’t hog memory—my 16GB machine handles it just fine.
One heads-up though: don’t stack steps with Turbo like you would with normal models. Adding more barely improves quality and just wastes time. Also, keep CFG low—crank it up and you’ll get blown-out colors that look dead. Stick to 1024 resolution for output, and upscale separately if you need bigger images. Once you’re happy with the composition, lock the seed, tweak one prompt at a time, and the layout won’t go haywire. The best part? Prompts and outputs stay local—no uploads, no billing.