ModelOption ✒️🅝🅐🅘

By default, GenerateNAID runs nai-diffusion-4-5-full. If that's what you want - and for most people it is - you never need this node. ModelOptionNAID exists for the days you want a different model without rebuilding your whole workflow: a curated/preview variant that costs less Anlas, an older checkpoint for a specific look, or the demo model to test a prompt for near-free.

It's one of the pack's "option" nodes, which means its whole job is to set a field on a settings dict and pass it along. You connect its NAID_OPTION output into the option input on GenerateNAID - and because it has its own option input, you can chain it with the other option nodes so a single Generate call is both img2img and model-switched.

The one input that matters is model, an enum. The current lineup tops out with the V4 family the README calls out explicitly:

• nai-diffusion-4-curated-preview and nai-diffusion-4-full
• nai-diffusion-4-5-curated and nai-diffusion-4-5-full - the newest, with 4-5-full being the pack's default

Below those sit the older generation the dropdown still carries - safe-diffusion (the free demo model), the original nai-diffusion, nai-diffusion-furry, nai-diffusion-2, nai-diffusion-furry-3, and nai-diffusion-3. "Curated" is NovelAI's cheaper, lighter variant; it produces slightly different (and usually faster, cheaper) output than the full model, so it's the natural choice for iteration. Don't assume the node defaults to the newest thing - a cached copy of the schema may show nai-diffusion-4-full as its default, while the current code ships nai-diffusion-4-5-full. If your dropdown's first entry doesn't match, update the pack and re-read.

A couple of interactions to know. The node works with the other options, but the action nodes (img2img, inpainting) decide their own behavior; ModelOption only changes which model runs that action. And if you're doing inpainting on a non-v2 model, the pack silently appends -inpainting to the model id it sends - so you don't pick that variant yourself. Vibe transfer, character reference, all of it runs under whatever model you select here.

Install is the pack's standard path - ComfyUI Manager → search "ComfyUI_NAIDGenerator", or:

cd ComfyUI/custom_nodes
git clone https://github.com/bedovyy/ComfyUI_NAIDGenerator

then restart. You still need the NAI_ACCESS_TOKEN in a .env in the ComfyUI root folder or the Generate node downstream will 401. No model files download - the whole point is NovelAI's servers hold the weights and your Anlas buys access to them.