How does NovelAI's tag weighting syntax work?

NAI uses curly braces to increase a tag's strength and square brackets to decrease it. Each pair of braces multiplies the effect by about 1.05, so {{tag}} is stronger than {tag}, and [tag] weakens it. This differs from the parenthesis-with-number syntax used by AUTOMATIC1111.

What quality tags should I use with NAI Diffusion V3?

V3 responds well to its own booster set, typically best quality, amazing quality, very aesthetic, and absurdres placed at the front. NAI also offers Quality Tags and Undesired Content presets in its UI that prepend these automatically.

What sampler and steps work best for NAI V3?

Euler and Euler Ancestral are reliable defaults, with DPM++ 2M and DPM++ SDE also popular. Around 28 steps is a strong sweet spot; going much higher rarely improves V3 output and just costs more Anlas.

What is the Undesired Content field?

It is NovelAI's name for the negative prompt — tags you want the model to avoid. NAI provides preset levels (Light, Heavy) that fill in common negatives like lowres, bad anatomy, and worst quality, and you can add your own on top.

NovelAI Diffusion Prompt Guide

NovelAI Diffusion V3 prompting

NovelAI Diffusion V3 is one of the strongest anime-and-illustration models, and like its imageboard-trained peers it speaks in tags. What sets NAI apart is its distinctive weighting syntax and its built-in quality and Undesired Content presets. Learn the handful of conventions below and you can drive it precisely.

How it works

NAI reads danbooru-style tags and applies emphasis through brackets rather than numbers. Curly braces strengthen a tag — {tag} is stronger, {{tag}} stronger still — and square brackets weaken it. Each layer adjusts the weight by roughly five percent, so you tune emphasis by stacking braces. At the front of the prompt go quality boosters (best quality, amazing quality, very aesthetic, absurdres), which bias V3 toward its cleanest outputs. Negatives live in the Undesired Content field, with preset levels that fill in common unwanted tags. This guide builds a correct positive prompt and shows the exact weighting syntax plus recommended sampler and step settings.

Tips and notes

Stack braces, don’t guess numbers. Want a stronger color? {{blue_eyes}}, not a numeric weight — NAI ignores AUTOMATIC1111-style (tag:1.4).
Use the presets, then refine. Turn on Quality Tags and a sensible Undesired Content level, then layer your own tags on top.
28 steps is plenty. V3 converges quickly; spending Anlas on 50+ steps rarely pays off.
Euler a for variety, DPM++ for consistency. Switch samplers when you want more or less variation between seeds.