Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-4A-NovelAI

ComfyUI Extension: ComfyUI-4A-NovelAI

Repo Name

ComfyUI-4A-NovelAI

Author
tsukino4a (Account age: 0 days)
Nodes
View all nodes(1)
Latest Updated
2026-11-03
Github Stars
0.01K

How to Install ComfyUI-4A-NovelAI

Install this extension via the ComfyUI Manager by searching for ComfyUI-4A-NovelAI
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-4A-NovelAI in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

ComfyUI-4A-NovelAI Description

ComfyUI-4A-NovelAI brings NovelAI image generation into ComfyUI workflows, with prompt and character controls, image-to-image editing, inpainting, V4.5 references, metadata reuse, and original-file saving.

ComfyUI-4A-NovelAI Introduction

ComfyUI-4A-NovelAI lets you create NovelAI images inside a ComfyUI workflow. It supports NovelAI Diffusion V4.5 and later only and brings prompt writing, multi-character scenes, image editing, references, and metadata reuse into connected ComfyUI nodes.

If you already use ComfyUI to organize your art process, this extension helps you keep NovelAI generation in the same workspace. You can write prompts with NovelAI’s own syntax, position characters in the frame, generate from an existing image, and save NovelAI’s original output without converting it again.

The extension is an unofficial community project and is not affiliated with NovelAI. Generation, Vibe encoding, and reference features may consume Anlas under your account’s current API rules. You use your own NovelAI Persistent API Token.

ComfyUI-4A-NovelAI overview

How ComfyUI-4A-NovelAI Works

Think of a workflow as a set of connected workstations. Each node prepares one part of the image request, and NAI Sampler sends the assembled request to NovelAI:

  1. NAI Model Loader chooses the NovelAI model.
  2. NAI Prompt Scheduler supplies your positive and negative prompts, character cards, and image dimensions.
  3. Optional nodes add an input image, painted mask, or reference resources.
  4. NAI Sampler requests the image and returns both a ComfyUI image and NovelAI’s original image data.
  5. NAI Original Image Saver writes the original PNG or lossless WebP to ComfyUI’s output folder without re-encoding it.

For a first text-to-image workflow, connect the Model Loader and the Scheduler to the Sampler, then connect the Sampler’s NAI_RESULT output to the Original Image Saver. The basic example workflow provides this arrangement. Select a model, configure your token, enter a prompt, and queue the workflow.

NAI Image Input changes the type of generation according to what you provide—not simply whether a mask socket is connected:

What you provideWhat happens
No source imageText-to-image: NovelAI creates an image from your prompts.
A source image with an empty or unpainted maskImage-to-image: NovelAI works from the existing image. An empty mask may remain connected.
A source image with a painted maskInpainting: NovelAI edits the marked area.

When you use a source image, NAI Image Input determines the effective generation size from that image’s aspect ratio. This helps prevent an unrelated size setting on the Sampler from stretching your source.

ComfyUI-4A-NovelAI Features

Native prompt writing and character placement

NAI Prompt Scheduler keeps NovelAI’s {}, [], and N::text:: prompt-weighting syntax intact. In its editor, selecting text and typing {, [, or : adds the matching closing character. Nested and numeric weights are also colored to make longer prompts easier to read.

You can switch quality, character, action, scene, and character-card fields between positive and negative text in place. Distinct colors help you see which side of the prompt you are editing without maintaining two large, separate layouts.

For scenes with several people, add ordered character cards up to the selected model’s limit. Leave placement automatic, or arrange characters on a canvas shaped like your output image. Manual placements are stored as normalized x/y positions, so they describe locations within the frame rather than fixed pixel coordinates.

NAI Prompt Scheduler

Image sizes and batch runs

The Scheduler offers Small, Normal, Large, and Wallpaper size tiers while preserving the current aspect ratio. You can also enter a custom width and height. The character-positioning canvas follows the effective image dimensions, helping you plan a composition that matches the generated frame.

Size tierPortraitSquareLandscape
Small512 × 768640 × 640768 × 512
Normal832 × 12161024 × 10241216 × 832
Large1024 × 15361472 × 14721536 × 1024
Wallpaper1088 × 1920—1920 × 1088

Custom dimensions range from 64 to 2048 pixels, in 64-pixel steps.

To repeat a fixed prompt, set a start index and task count, then use the Scheduler’s Batch Run button. It adds all jobs to the ComfyUI queue immediately; ComfyUI still processes them one at a time. Changing prompts randomly or sequentially between jobs requires the optional 4A Prompt Manager integration described below.

Optional 4A Prompt Manager Wildcards

With ComfyUI-4A-Prompt-Manager alongside this extension, the Scheduler can read its folder-backed TXT and JSON-card prompt library. A token such as __folder__ or __path/file__ can draw from that library, letting you switch prompt elements randomly or sequentially. Sequential mode can count the available leaf choices and queue the batch at once.

Wildcard selection has its own seed, separate from the image seed on NAI Sampler. If sampling-data apply is enabled, the connected Sampler can use supported sampling fields from the chosen card at run time. 4A Prompt Manager can also send native NovelAI prompts, character cards, and supported metadata fields to every matching NAI node in a workflow.

Only the plain __wildcard__ form is supported. NovelAI’s braces remain prompt-weighting syntax; they are not interpreted as random choices. Without 4A Prompt Manager, wildcard tokens are not expanded, but the normal generation nodes and fixed-prompt batches still work.

4A Prompt Manager Wildcard integration

Image-to-image and inpainting controls

NAI Image Input manages the source image, mask, strength, noise, and output size for image-to-image and inpainting. You can retain approximately the source image’s pixel count or scale it to a selected size tier while keeping its aspect ratio.

An Upscale 1.5 option follows NovelAI’s official Enhance sizing rule: multiply each side by 1.5, round upward to a 64-pixel step, and scale back if either side would exceed 2048 pixels. Upscale 1.5 and the pixel-budget option cannot be used together.

NAI Sampler can request PNG or lossless WebP and can use NovelAI’s streaming endpoint for live progress previews.

NAI image generation modes

Vibe and Precise Reference cards

NAI Reference Resources lets you drop in images or .naiv4vibe files and manage them as cards. Choose one of two mutually exclusive approaches:

  • Vibe uses Vibe resources to influence the generated image.
  • Precise Reference cards can target a character, a style, or both character and style.

When a matching local Vibe encoding exists, image Vibes can reuse it. A missing encoding is never purchased silently: encoding requires the explicit Encode and save action and your confirmation.

These reference resources work with the configured V4.5 models. V5 ignores connected Vibe and Precise Reference resources, without displaying an error; remove or bypass those resources when planning a V5 workflow.

Vibe and Precise Reference cards

Metadata inspection and reuse

NAI Meta Loader displays information from NovelAI images as cards, including the model, generation settings, positive and negative prompts, and individual characters. You can send a particular model, setting, or prompt to your workflow, or append a character. Its combined prompt action replaces the full prompt and character set.

NAI Meta Apply lets you choose which fields to reuse from a dropped or selected NovelAI image: positive prompt, negative prompt, characters, character append mode, model, generation settings, and seed. The Loader and Meta Apply send values to every matching NAI node in the workflow; you do not need to select a target node first.

Some choices are deliberately limited to protect your current composition:

  • Applying model and generation settings does not overwrite resolution.
  • Applying only the positive prompt preserves the Scheduler’s character, action, and scene fields.
  • For a non-NovelAI image, the tools can inspect positive and negative prompt text from 4A Prompt Manager, A1111, or native ComfyUI metadata, in that order. They do not treat that image’s model or generation settings as NovelAI settings.

NAI metadata tools

Token, account usage, and original image saving

Configure your Persistent API Token through Settings → 4A NovelAI → Account → Token & Anlas or NAI Usage Monitor. The token goes to your local ComfyUI backend and is stored outside workflows at:

<ComfyUI user>/ComfyUI-4A-NovelAI/credentials.json

It is not written into workflows, generated-image metadata, browser settings, or plugin logs. NAI Usage Monitor displays your subscription tier, included and purchased Anlas, and available V5 allowance.

NAI Original Image Saver saves the PNG or lossless WebP returned by NovelAI directly to ComfyUI output, without re-encoding. This preserves the original file and its provider metadata.

NAI usage monitor

ComfyUI-4A-NovelAI Models

Choose a model in NAI Model Loader. The available selections are:

Model selectionWhat to keep in mind
V5 FullA V5 option. Connected Vibe and Precise Reference resources are ignored.
V5 CuratedA V5 option with the same reference-resource limitation. Its inpainting route follows V4.5 Curated inpainting and has a six-character limit in that mode.
V4.5 FullA V4.5 option that can use the configured Vibe or Precise Reference cards.
V4.5 CuratedA V4.5 option that can use the configured Vibe or Precise Reference cards.

If references are essential to your idea, choose a V4.5 model. If you choose V5, do not expect connected reference cards to affect the result. The supplied documentation does not specify further visual-style or performance differences between Full and Curated, so comparing their results with your own prompts is the most useful way to choose between them.

What's New with ComfyUI-4A-NovelAI

Current release: 1.1.3

  • The metadata tools can read positive and negative prompts from native ComfyUI metadata when a non-NovelAI image has neither 4A Prompt Manager nor A1111 metadata. This gives you another way to recover prompt text from images you want to revisit.
  • Support for the newer ComfyUI assets panel allows images dragged from output subfolders to load correctly.

Version 1.1.2

  • 4A Prompt Manager sends and the metadata tools reach every matching NAI node in the workflow without target selection. This is useful when your canvas contains more than one Scheduler or Sampler.
  • Wildcard expansion converts ComfyUI prompts using 4A Prompt Manager’s weight converter, helping prompts selected from its library work with NovelAI’s prompt syntax.

Troubleshooting ComfyUI-4A-NovelAI

A __wildcard__ token appears unchanged

  1. Check whether ComfyUI-4A-Prompt-Manager is available alongside this extension. Wildcard expansion depends on it.
  2. Use the plain __wildcard__ form, pointing to a library entry such as __folder__ or __path/file__.
  3. Do not use {...} as a random-choice expression; braces have a native NovelAI prompt meaning.
  4. Without 4A Prompt Manager, replace the token with fixed prompt text.

A source image is edited when you expected inpainting

  1. Check the mask supplied to NAI Image Input.
  2. Paint the area you want to inpaint. An empty or unpainted mask keeps the request in image-to-image mode, even when the mask socket is connected.
  3. Queue the workflow again with the painted mask.

NAI image generation modes

My reference cards do not change a V5 image

  1. Check the selection in NAI Model Loader.
  2. If it is V5 Full or V5 Curated, switch to a V4.5 option when you want to use Vibe or Precise Reference cards.
  3. Queue the workflow again. V5 silently ignores connected reference resources.

My image-to-image output is not using the Sampler’s dimensions

When a source image is present, NAI Image Input owns the effective size so it can preserve the source aspect ratio. Set its size behavior to the pixel budget, approximate source pixel count, or Upscale 1.5 option you want. The Sampler’s unrelated dimensions will not simply replace that size.

Metadata did not change my resolution or every prompt field

Check which metadata action you used. Applying model and generation settings intentionally leaves resolution alone. Applying only the positive prompt also preserves separate character, action, and scene fields. Use the combined prompt action when you want the complete prompt and character set replaced. For non-NovelAI images, only positive and negative prompt text is interpreted—not their model and generation settings.

Frequently asked questions

Do I need 4A Prompt Manager to generate images?
No. The Model Loader, Scheduler, Sampler, and other generation nodes work without it. You need the optional integration for its wildcard library, prompt switching, and metadata bridge.

Will an empty connected mask trigger inpainting?
No. With a source image, an empty or unpainted mask means image-to-image; a painted mask means inpainting.

Does a batch run generate every image at the same time?
No. Batch Run adds all jobs to ComfyUI’s queue immediately, and ComfyUI processes them one by one.

Will encoding a missing Vibe happen automatically?
No. You must choose Encode and save and confirm the action.

Does saving through NAI Original Image Saver convert the image?
No. It writes NovelAI’s returned PNG or lossless WebP bytes without re-encoding them.

Are failed requests always retried?
No. Automatic retries are limited to transient failures for single-image text generation within the Normal pixel budget, with no more than 28 steps and no paid reference features. Large or Wallpaper requests, image-to-image, inpainting, more than four Vibes, Precise Reference, and Vibe encoding are sent only once.

Learn More about ComfyUI-4A-NovelAI

  • Basic example workflow: Load this in ComfyUI to see the essential Model Loader → Prompt Scheduler → Sampler → Original Image Saver connections. Choose a model and configure your token before queueing it.
  • ComfyUI-4A-Prompt-Manager: Explore the optional folder-backed prompt library if you want reusable Wildcards, sequential prompt tracks, or its metadata bridge.
  • Chinese video guide on Bilibili: A visual guide if you prefer to follow the interface on screen; the narration is in Chinese.
  • Chinese-language documentation: An alternative version of the extension documentation.

ComfyUI-4A-NovelAI Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

ComfyUI-4A-NovelAI detailed guide | ComfyUI