New in Nook: Train LLMs
Give it a few examples and Nook fine-tunes a model on your own GPU, completely privately.
We taught Qwen3 4B to talk like a pirate with 32 examples: 14 seconds of training on an RTX 4060. Then it answered questions it never saw.
usenook.ai
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.
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.
9/9
Download and update your Nook and experience the best platform ever created to experiment with local and private AI.
usenook.ai
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.
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.
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.
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.
Tomorrow, Nook will unveil one of the best tools that's ever been built for generating images and videos locally.
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.
8/8 Got a different number on your machine? Send it with your setup and we'll add reader results to the note.
Read the first note: usenook.ai/labs/introducing-…
RSS: usenook.ai/labs/feed.xml
