ComfyUI

ComfyUI is an open-source, node-based interface and inference engine for constructing and running generative-artificial-intelligence workflows. It was written by the pseudonymous developer comfyanonymous and other contributors and is maintained in the Comfy-Org organization on GitHub.¹²

ComfyUI is not itself a generative model. It provides the graph editor, execution engine, model loaders, queue, and interfaces used to connect models and operations into reproducible workflows. The model weights used by a graph remain separate software and data artifacts.

Workflow model

A ComfyUI workflow is a graph of connected nodes. Each node performs a task, exposes typed inputs and outputs, and passes its results to other nodes through links. A graph can therefore express a process such as loading a model, encoding a prompt, sampling a latent representation, decoding an image, and saving the result without fixing those steps behind a single-purpose interface.³⁴

Workflows can be saved as human-readable JSON. ComfyUI also embeds workflow information in supported generated files, allowing a user to reopen the graph that produced an output. When a graph is run again, the execution engine can reuse unchanged results and execute only the portions affected by changed inputs.³²

The local server accepts workflows through an HTTP API, places them in an execution queue, and uses a WebSocket connection for progress messages. This allows the visual interface and automated production pipelines to use the same graph execution model.⁵

Models and media

ComfyUI became widely associated with diffusion-model image workflows and supports families including Stable Diffusion and Stable Diffusion XL. Its scope is broader than one model family: the project documents workflows for image generation and editing, video, audio, and 3D content. Exact model support changes as core nodes and templates are updated.²

The core application can run locally and does not require Comfy’s hosted service. Official distribution options include desktop builds, a Windows portable package, and manual installation on Windows, Linux, and macOS. Comfy Cloud is a separate hosted way to execute compatible workflows; optional API nodes can also call external model providers.²⁶

Extensibility and operational considerations

The base installation includes officially maintained Comfy Core nodes. Community authors can add custom nodes, which may implement new model integrations, processing operations, or interface extensions. Custom nodes can bring their own Python dependencies, and incompatible dependency pins can break other extensions in the same environment.⁷

Because custom nodes are executable extensions rather than inert workflow data, their source and provenance matter. ComfyUI’s registry standards prohibit obfuscated code, runtime package installation, and uses of eval or exec that can enable arbitrary-code-execution vulnerabilities.⁸ A shared workflow may therefore be portable as JSON while still requiring particular model files and trusted node packages to run correctly.

Use with Rina

At Rio Maker Space’s Arduino Day 2025, Rodrigo Laneth presented ComfyUI as the visual counterpart to the Rina language-model demonstration. The combined lesson introduced attendees with no previous AI-character-design or system-prompting experience to both conversational character design and the generation of visual character assets.

Laneth self-hosted both the Rina services and ComfyUI on the same dedicated cloud GPU instance. This was a deployment of the open-source ComfyUI server on infrastructure controlled by Laneth, not a use of the Comfy Cloud service. Rina’s language-model chat and the ComfyUI image-generation workflows were separate workloads sharing that compute infrastructure.

References

Footnotes

1. ComfyUI official documentation. ↩

2. ComfyUI source repository. ↩ ↩² ↩³ ↩⁴

3. Workflow, ComfyUI documentation. ↩ ↩²

4. Nodes, ComfyUI documentation. ↩

5. ComfyUI server routes, ComfyUI documentation. ↩

6. Manual installation, ComfyUI documentation. ↩

7. Custom nodes, ComfyUI documentation. ↩

8. Registry standards, ComfyUI documentation. ↩