7/9 Prefer nodes? The same recipe flips into a graph, like ComfyUI. Got a ComfyUI workflow? Open it and Studio turns its nodes into steps, Wan video workflows included, and tells you what it left out.
1
4
52
8/9 The models: SD Turbo (fits a 6 GB card), SDXL Turbo and Z-Image Turbo for pictures, Real-ESRGAN to make them bigger, Wan for video. Each says whether your card can run it before you download. Your own LoRAs go in an add-ons folder.
1
4
35
9/9 Download and update your Nook and experience the best platform ever created to experiment with local and private AI. usenook.ai
5
20
6/9 New: video. "Video from words" makes a short clip with the Wan models, right in the same recipe. On an 8 GB card like an RTX 4060, a 2-second Wan 2.1 clip takes about five minutes. The catalog also has Wan 2.2 and the 14B models, each marked for the cards it fits.
1
4
45
1/9 Meet Nook Studio: Generate and edit pictures, and now short videos, on your own graphics card. It's the power of ComfyUI without the wiring. A recipe is a few steps in plain words, and you can flip it into nodes whenever you want.
2
2
11
311
5/9 Add a step after a picture you already made, and only the new step runs. Change a setting, and only the steps from there on are made again. It's how ComfyUI skips nodes whose inputs didn't change, minus the graph.
1
3
36
4/9 Fix one part without touching the rest. Paint over it, say what goes there, and only that part is drawn again. Everything outside the paint stays exactly as it was, pixel for pixel.
1
4
35
3/9 Every step sits on a colourful taskbar at the bottom. Click one to add it, or drag it where you want it. Not sure where to start? Studio opens by asking: a fresh recipe, a ready-made one, or one you saved.
1
4
39
2/9 A recipe is steps, one after another: Imagine → Change it → Repaint a part → Make it bigger. Each step hands its pictures to the next, and you watch the engine paint as it works. Your prompts and pictures stay on your PC.
1
6
94
Tomorrow, Nook will unveil one of the best tools that's ever been built for generating images and videos locally.
9
306
1/8 Introducing Nook Labs. We build Nook, an app that runs AI on your own GPU. That means testing inference on the computers people actually own, constantly. Those results used to die in our commit messages. Now we publish them. 🧵
2
1
9
528
7/8 Next on the list: • What quantisation really costs (Q8 down to Q3 on a coding model) • What actually fits in 8 GB • Whisper tiny to large: speed vs accuracy • Apple silicon vs a PC at the same price What should we test first? Reply and tell us.
1
2
45
Where does your prompt go? Cloud AI: someone else's servers. Depending on the provider, it can be stored, reviewed or used for training. Nook: your own GPU. No account. No cloud. Pull the cable and it still works. Private inference, on your PC.
4
247
Share the music you make 🎛️ New Nooklet: Record your screen → A screen, a window or an area → Your mic + your computer's sound → Go live on Twitch, YouTube, Kick or any RTMP server → Keep a recording while you stream Encoded on your GPU. Recordings stay on your PC.
3
7
324