In ComfyUI, a custom node pack is a Python package placed inside ComfyUI/custom_nodes. When ComfyUI starts, it loads packages that export NODE_CLASS_MAPPINGS.

This repository is set up to teach and support the full lifecycle:

flowchart LR
local[Local node development] --> git[Git history]
git --> github[GitHub repository]
github --> registry[Comfy Registry publish]
registry --> manager[ComfyUI-Manager install/update]

Current nodes

Dustin Text Prefix

Category: Dustin Nodes/Text

Inputs:

text: source text.
prefix: text added before the source.
separator: text inserted between prefix and text.

Output:

text: combined string.

This first node is intentionally small so you can verify repository structure, loading behavior, and publishing flow before adding more advanced nodes.

Dustin Image Atlas

Category: Dustin Nodes/Image

Inputs:

images: a batch of square images.
max_width: maximum atlas width in pixels.
max_height: maximum atlas height in pixels.
padding: empty pixels inserted between tiles.

Outputs:

atlas_image: one stitched atlas image.
atlas_metadata: JSON text containing each tile's index and position.

Behavior:

Places images from left to right.
Wraps to a new row when the next image would exceed max_width.
Raises an error instead of resizing or dropping images when the atlas would exceed max_height.

Dustin Marble Generate World

Category: Dustin Nodes/Marble

Calls the World Labs Marble API to generate a 3D world, then polls until the long-running operation completes (typically about 5 minutes).

Authentication:

Set api_key on the node, or
Set the WLT_API_KEY environment variable before starting ComfyUI.

Inputs:

prompt_type:
- text: text-to-world (text_prompt required).
- image_url: image-to-world from a public URL (image_url required).
- image_tensor: image-to-world from a ComfyUI IMAGE input (image required).
- multi_image_json: multi-image from a JSON array (multi_image_json or builder node output).
- video_url: video-to-world from a public URL (video_url required).
- video_file: local video file (video_path or media_asset_id required).
text_prompt: text guidance (required for text; optional for other modes).
model: marble-1.1, marble-1.1-plus, marble-1.0, or marble-1.0-draft.
display_name: optional world name (max 64 characters).
seed: random seed (-1 omits seed from the request).
is_pano: mark the input image as a panorama (image modes only).
reconstruct_images: multi-image reconstruction mode (up to 8 images).
disable_recaption: send text_prompt as-is without automatic recaptioning.
poll_interval_seconds / max_wait_seconds: operation polling controls.

Outputs:

world_json: full world payload from the completed operation.
world_id: generated world ID.
marble_url: link to open the world in Marble.
asset_urls_json: thumbnail, panorama, mesh, and Gaussian splat download URLs.

Recommended image formats for URL / tensor modes: jpg, jpeg, png, webp. Recommended video formats: mp4, mov, mkv.

Dustin Marble Build Multi-Image Prompt

Category: Dustin Nodes/Marble

Uploads an IMAGE batch and builds multi_image_json for the generate node.

Inputs:

images: square scene photos as an image batch.
azimuths: comma-separated degrees, one per image (example: 0,90,180,270).
upload_mode: media_asset (recommended) or data_base64.

Output:

multi_image_json: JSON array for Dustin Marble Generate World (prompt_type = multi_image_json).

Dustin Marble Generate Multi-Image World

Category: Dustin Nodes/Marble

All-in-one multi-image generation: uploads batch images, starts generation, polls until done.

Dustin Marble Upload Video

Category: Dustin Nodes/Marble

Uploads a local video via Marble media-assets:prepare_upload and returns media_asset_id.

Inputs:

video_path: absolute or relative path to mp4, mov, mkv, etc.

Dustin Marble Generate Video World

Category: Dustin Nodes/Marble

Video-to-world generation from either a public video_url or a local file / uploaded media_asset_id.

Dustin Marble Preview Thumbnail

Category: Dustin Nodes/Marble

Downloads the world thumbnail and shows it in the ComfyUI preview panel.

Inputs (either one):

asset_urls_json from a generate node, or
image_url with a direct thumbnail URL.

Outputs:

image: ComfyUI IMAGE tensor for downstream nodes.

Dustin Marble Preview Panorama

Category: Dustin Nodes/Marble

Downloads the world panorama (pano_url) and shows it in the ComfyUI preview panel as a flat image.

Inputs (either one):

asset_urls_json from a generate node (recommended), or full world_json (also supported),
image_url with a direct panorama URL.

Outputs:

image: ComfyUI IMAGE tensor (panoramas can be very wide).

Dustin Marble Pano 360 Viewer

Category: Dustin Nodes/Marble

Interactive equirectangular panorama viewer embedded in the node. After you queue the workflow, drag inside the node to look around and use the mouse wheel to zoom (similar to a 360° web viewer).

Uses a DOM widget with an embedded viewer page (js/pano_embed.html) so pointer events are not stolen by the workflow canvas (works in both Node 1.0 and Node 2.0 UIs). Does not show the default flat ComfyUI image preview at the bottom of the node. Pannellum is bundled under js/vendor/pannellum/ (no CDN).

Recommended workflow:

Dustin Marble Generate World → asset_urls_json → Dustin Marble Pano 360 Viewer

Inputs (at least one):

asset_urls_json from a generate node (uses pano_url),
image_url with a direct panorama URL, or
image from an upstream IMAGE output.

Options:

max_preview_width: downscales very large panoramas before preview (default 4096) to avoid WebGL limits.

Outputs:

image: ComfyUI IMAGE tensor for downstream nodes (Save Image, etc.).

Notes:

Panoramas are downloaded on the server first (avoids CORS issues with Marble CDN URLs).
Use Preview Panorama for a quick flat preview; use Pano 360 Viewer for interactive look-around.

Dustin Image Atlas Extract

Category: Dustin Nodes/Image

Inputs:

atlas_image: the stitched atlas image from Dustin Image Atlas.
atlas_metadata: JSON metadata from Dustin Image Atlas.
index: the tile number to extract, starting from 0.

Output:

image: the cropped tile for the selected index.

Local installation

Copy or clone this repository into your ComfyUI/custom_nodes folder:

cd path\to\ComfyUI\custom_nodes
git clone https://github.com/YOUR_GITHUB_USERNAME/dustin-comfyui-nodes.git

Then restart ComfyUI and search for Dustin to find all nodes in this pack.

Repository layout

__init__.py: ComfyUI package entry point.
nodes/: node implementations and exports.
pyproject.toml: Comfy Registry metadata.
requirements.txt: runtime Python dependencies.
docs/full-flow.zh-CN.md: step-by-step Chinese guide for Git, GitHub, publishing, and updates.

Publishing outline

ComfyUI-Manager now uses Comfy Registry as the main distribution path.

Create a GitHub repository.
Push this code to GitHub.
Create a publisher on registry.comfy.org.
Generate a Registry API key.
Fill in the real repository URL and PublisherId in pyproject.toml.
Run comfy node publish.

Once a version is published, it cannot be edited. You must bump the version and publish again for every fix or new node.

Versioning

This project uses semantic versioning:

PATCH, for example 0.1.1: bug fix or documentation fix.
MINOR, for example 0.2.0: new node or backward-compatible feature.
MAJOR, for example 1.0.0: breaking change.

Adding a new node later

Create a new node class inside nodes/.
Export it from nodes/__init__.py.
Add it to NODE_CLASS_MAPPINGS and NODE_DISPLAY_NAME_MAPPINGS.
Document it here.
Test it locally in ComfyUI.
Bump version in pyproject.toml.
Push and publish the new version.

Notes

Most nodes use only the Python standard library. The Marble node also requires requests (see requirements.txt).