NovelAI: Adjusting prompt weights with brackets

NovelAI Try it →Watch full video →NovelAI · Oct 2024

Demo summary

The tutorial demonstrates using curly brackets to strengthen tags and square brackets to weaken them, showing the visual difference in the resulting image's hair texture.

Step-by-step

Click the seed button to lock the current image generation
Surround a tag with {curly brackets} to strengthen its effect
Add multiple layers of brackets to increase the strength further
Surround a tag with [square brackets] to weaken its effect
Ensure the number of opening and closing brackets match on both sides

Options

Use square brackets for tags that appear too extreme, such as makeup

Watch out for

Using too many brackets can cause strange visual effects in the image
Brackets must be balanced on both sides unless the tag is at the very end of the prompt

Tips

Lock the seed before adjusting weights so you can see changes without the entire image shifting

Highlights

“you can get a more granular control over what the AI will focus on”

All demos from “NovelAI Image Generation Intro Tutorial”

0:240:17Basic image generation in NovelAIThe user demonstrates entering simple text prompts like 'flower field' and 'sunset' into NovelAI Diffusion to generate an anime-style landscape.NovelAI· Text to Image
1:180:20Using tag suggestions and prompt barThe video shows the NovelAI interface suggesting specific tags as the user types and demonstrates the prompt length bar filling up.NovelAI· AI Image Generator
1:380:31Character generation with specific tagsA character is generated using specific descriptive tags like '1girl', 'school uniform', and 'green eyes' to show how NovelAI handles detailed prompts.NovelAI· AI Person Generator
2:090:49Adjusting prompt weights with bracketsCurrentThe tutorial demonstrates using curly brackets to strengthen tags and square brackets to weaken them, showing the visual difference in the resulting image's hair texture.NovelAI· AI Image Generator
2:580:23Removing elements with Undesired ContentThe user demonstrates how to use the 'Undesired Content' field to remove specific elements, such as flower petals, from a generated image.NovelAI· AI Object Remover/Eraser
4:320:21Recovering prompts via drag and dropThe video shows how to recover the original prompt and seed settings by dragging a previously saved NovelAI image back into the web interface.NovelAI· AI Image Generator
Watch “NovelAI Image Generation Intro Tutorial” →

AI Image Generator

1:180:20Using tag suggestions and prompt barThe video shows the NovelAI interface suggesting specific tags as the user types and demonstrates the prompt length bar filling up.NovelAI
2:090:49Adjusting prompt weights with bracketsCurrentThe tutorial demonstrates using curly brackets to strengthen tags and square brackets to weaken them, showing the visual difference in the resulting image's hair texture.NovelAI
4:320:21Recovering prompts via drag and dropThe video shows how to recover the original prompt and seed settings by dragging a previously saved NovelAI image back into the web interface.NovelAI
0:430:58Generate images using the Quick Start Gallery in NovelAIThe user demonstrates how to select an existing image from the NovelAI gallery to automatically copy its prompts and settings for a new generation.ぷろもでAI研究所【生成AI解説】
3:111:12Configuring V4.5 Full model and quality tagsThe demonstrator selects the 'V4.5 Full' model, applies quality tag presets, and adjusts manual settings like Steps and Prompt Guidance.ぷろもでAI研究所【生成AI解説】
7:051:32Using Negative Prompts and Undesired Content presetsThe video shows how to access the 'Undesired Content' field to add negative prompts and select presets to fix common AI artifacts like incorrect hand anatomy.ぷろもでAI研究所【生成AI解説】
16:321:12Using Question Mark Categories for AI CorrectionThe creator shows a specific technique using a question mark in category headers (e.g., 'Legs?') to provide the AI with negative constraints or specific corrections for common mistakes.RentyAi
5:330:46Comparing NovelAI GLM 4.6 vs Xialong speedThe creator switches back to the older GLM 4.6 model to demonstrate the visible difference in generation latency and 'hitching' compared to the new Xialong model.RentyAi
11:371:36Exploring Xialong model settings and samplersThe video shows the NovelAI interface for the Xialong model, highlighting the simplified preset options and the reduced number of available samplers.RentyAi