> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bizyair.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Online ComfyUI

> BizyAir Online ComfyUI deploys open-source ComfyUI in the cloud with workflow publishing and community sharing.

Online ComfyUI deploys the full capabilities of open-source ComfyUI in the cloud and extends it with platform features such as workflow publishing, API integration, and community sharing. Think of it as an all-in-one product: "ComfyUI + model library + one-click API + workflow marketplace".

## Core Capabilities

| Capability                | Description                                                                           |
| ------------------------- | ------------------------------------------------------------------------------------- |
| Node-based editing        | Drag and connect nodes to orchestrate image / video / audio / 3D generation pipelines |
| Ready-to-use models       | Hundreds of open-source and proprietary models built in — no local downloads needed   |
| Workspace management      | Personal workflows, generation history, and resources all in one place                |
| One-click publishing      | Publish workflows as APIs callable by external apps or agents                         |
| Community sharing         | Browse, fork, and favorite community workflows to bring ideas to life faster          |
| Usage-based billing       | Elastic cloud GPUs billed by runtime — no extra fees                                  |
| Architecture optimization | Reduced generation time through compute optimization and self-deployed models         |

## Use Cases

* **Rapid prototyping**: Validate generative AI pipelines on devices without a GPU
* **API-as-a-service**: Turn designer-tuned workflows into live endpoints
* **Teaching demos**: Demonstrate node-based pipelines in class without per-student installs
* **Creative sharing**: Circulate workflows as creative assets in the community for remixing

***

## Workspace

The workspace is the default page after entering Online ComfyUI. It combines "My Workflows", "New Workflow", "Folder categorization", "New Folder", and "Filter & Sort".

### Layout Overview

The workspace is divided into a right-side function area, an upper content area, and a lower content area.

| Area                     | Main content                                       |
| ------------------------ | -------------------------------------------------- |
| Right-side function area | New workflow / folder, filter & sort               |
| Upper content area       | Folders (for archiving and categorizing workflows) |
| Lower content area       | Workflow history (including date and name)         |

***

## Workflow Editing

The editor is the core canvas for node-based workflows and works much like local ComfyUI.

### Node Editor

#### Canvas Operations

| Action      | Mouse / shortcut                      | Effect                          |
| ----------- | ------------------------------------- | ------------------------------- |
| Pan canvas  | Right-drag / `Space + left-click`     | Move the view                   |
| Zoom        | Scroll wheel / `Cmd/Ctrl + +/-`       | Zoom in / out                   |
| Add node    | Double-click empty area               | Open the node search panel      |
| Select node | Left-click / `Shift` for multi-select | Prepare for action              |
| Copy node   | `Cmd/Ctrl + C` → `V`                  | Clone on canvas                 |
| Delete node | Select then press `Delete`            | Remove node and its connections |
| Disconnect  | Right-click a connection → Remove     | Remove only the edge            |
| Box select  | Left-drag on empty area               | Batch select and move           |
| Tidy layout | `Cmd/Ctrl + Shift + L`                | Auto-rearrange nodes            |

#### Adding Nodes

1. Double-click an empty area of the canvas.
2. Type a node name in the search box that appears (e.g., `Load Checkpoint`, `KSampler`, `VAE Decode`).
3. Click or press Enter — the node drops at the current canvas position.
4. Once placed, click and drag from its ports to connect to matching or compatible ports on the next node.

#### Parameter Configuration

* Clicking a node field opens the corresponding input control (dropdown, text, slider, etc.).
* Some nodes support both "fixed value" and "reference input" modes: fixed values are set at edit time; reference inputs can be exposed as API parameters when the workflow is published.
* Hovering over a field shows a tooltip.

### Node Library

Online ComfyUI's node library is organized into the following categories:

| Category      | Examples                                                                         |
| ------------- | -------------------------------------------------------------------------------- |
| Loader        | `Checkpoint Loader`, `LoRA Loader`, `VAE Loader`, `ControlNet Loader`            |
| Conditioning  | `CLIP Text Encode`, `Conditioning Combine`                                       |
| Sampling      | `KSampler`, `KSampler (Advanced)`, `SamplerCustom`                               |
| Latent        | `Empty Latent Image`, `Latent Upscale`                                           |
| Image / Video | `VAE Decode`, `Image Upscale`, `Video Linear CF`                                 |
| Custom        | Pre-installed extensions such as `ComfyUI-Impact-Pack`, `ComfyUI-ControlNet-Aux` |

<Note>
  If a node is missing, please go to [Feedback](https://forms.gle/Mo62qLuvUnmsChto9).
</Note>

### Models and Resources

When loading a model in a node, the dropdown prioritizes:

1. Official models from the **platform model library** (mounted directly from the cloud — no download needed).
2. Models and LoRAs **you have favorited or uploaded**.
3. Common model combinations **recommended by community workflows**.

When you select a platform library model, the node automatically reads the model's metadata (resolution tier, recommended sampler, etc.) as defaults.

### Running and Previewing

1. Finish configuring the workflow nodes and enter the relevant prompts.
2. Click **Run** (or `Cmd/Ctrl + Enter`).
3. A progress bar appears on the nodes — completed nodes turn green, running ones blue, and errored ones red.
4. Result nodes (e.g., `Save Image`) render thumbnails in real time; click to enlarge.
5. You can click **Stop** at any time during a run to cancel; time already run is still billed.
6. Each task carries a unique task ID (`requestId`) for later status and result queries.

<Note>
  For long videos or multi-step sampling, try a small-resolution dry run first before scaling up the output, to avoid wasting compute.
</Note>

The `data.status` in the response will show `Queuing` / `Preparing` / `Running` / `Success` / `Failed` / `Canceled`. See [API Reference](/api-guide/reference).

### Saving and Versioning

* **Manual save**: Click "Save" in the upper-left corner to overwrite the current workflow's draft.
* **Auto-save**: The workflow content is automatically saved after a successful run.
* **Version**: The platform only retains the most recent manually saved / auto-saved workflow content.

***

## Workflow Publishing

Once published, other users can invoke the workflow within the community.

### Pre-publish Checklist

Before clicking publish, confirm that:

* The workflow runs successfully (the latest draft run had no errors).
* Parameters that need external control are switched to "reference input", with names and default values defined.
* Output nodes are configured (e.g., `Save Image` path, `Save Video` filename rule).

### Publishing a Workflow

In the "Publish Workflow" panel:

| Field                | Description                                                        |
| -------------------- | ------------------------------------------------------------------ |
| Workflow name        | Field name used in API calls; lowercase English + underscores only |
| Version name         | e.g., `V1.0`, `V2.0`                                               |
| Base model           | Base model selection, e.g., `Qwen-Image`, `Z-Image`                |
| Cover image          | A visual introduction to the workflow                              |
| Workflow description | Helps users understand the workflow's function and type            |
| Public visibility    | Choose public or private                                           |

***

## Workflow Community

The workflow community is an open marketplace where users share, discover, and remix workflows. All public workflows appear here and can be searched, forked, favorited, and used.

### Browse and Discover

Workflows are organized by base model.

### Search and Filter

The search box on the right supports:

* **Keyword**: Matches workflow names
* **Sort**: Comprehensive / Recent / Most favorited / Most downloads / Most likes
* **Filter**: Base model / Runnable online

### Fork and Copy

1. On the workflow detail page, click `Run Workflow`.
2. Click `Run` in the upper-right corner.
3. After a successful run, you can find the workflow in your workspace.

### Favorite and Like

* **Like**: A one-time signal that affects the workflow's popularity ranking.
* **Favorite**: Adds to your personal collection for easy access later.
* **Follow author**: Get notified in the workspace "Activity" feed when the author publishes new work.

### Share and Download

* **Share**: Click `Share` to copy the workflow link.
* **Download**: In the right-side tab bar, click `Download` to download the workflow to your local machine.

***

## Task Lifecycle

Tasks for workflow runs go through the following stages; only the "Running" stage is billable:

| Stage                   | Billable                           | Description                                                    |
| ----------------------- | ---------------------------------- | -------------------------------------------------------------- |
| Queuing                 | Free                               | Task waiting in queue for GPU scheduling                       |
| Environment preparation | Free                               | Downloading model weights, pulling images, loading resources   |
| Running                 | Billable                           | Actual GPU inference time, billed at `5 BZ/sec` (`$0.005/sec`) |
| User interruption       | Billed for the portion already run | Successfully run nodes still incur charges                     |
| Platform interruption   | Free                               | Interruptions caused by the platform do not incur charges      |

<Note>
  Billing is based only on `inference_cost_time` (the inference portion); other stages are free. See [Pricing & Billing](/pricing).
</Note>