ERNIE-Image

ERNIE-Image Civitai ComfyUI Custom Nodes Practical Guide: 160+ Cloud AI Nodes Zero-GPU Workflow

Jul 2, 2026

ERNIE-Image Civitai ComfyUI Custom Nodes Practical Guide: 160+ Cloud AI Nodes Zero-GPU Workflow

ComfyUI is the most flexible workflow tool in the AI image generation space, but traditional ComfyUI workflows require a local GPU, model downloads, and significant disk space. Civitai ComfyUI custom nodes completely change this paradigm — you can now call cloud AI workflows directly within ComfyUI, with zero local hardware.

The civitai/civitai-comfy-nodes project provides approximately 160 auto-generated ComfyUI nodes that expose Civitai Orchestration API endpoints directly in ComfyUI graphs. ERNIE-Image, as a core ecosystem, can achieve zero-GPU-cost cloud text-to-image generation through these nodes.

This article walks you through configuring Civitai ComfyUI nodes from scratch, building ERNIE-Image cloud workflows, and diving deep into LoRA selectors, ControlNet, and the AIR URN system.

What Are Civitai ComfyUI Custom Nodes

Civitai ComfyUI custom nodes are a node pack installable through ComfyUI Manager, with the core function of converting all Civitai Orchestration API recipe endpoints into native ComfyUI nodes.

Key features:

  • Zero local GPU required: All computation runs on Civitai's cloud fleet
  • ~160 auto-generated nodes: Auto-generated from the Orchestration OpenAPI specification
  • Native ComfyUI type outputs: Nodes return IMAGE/VIDEO/AUDIO native types, seamlessly connecting to ComfyUI workflows
  • Pay-per-Buzz billing: Uses Civitai's Buzz credit system, pay for what you use

Installation Methods

ComfyUI Manager (Recommended)

  1. Open ComfyUI → Manager → Custom Nodes Manager
  2. Search for Civitai Comfy Nodes (publisher: civitai)
  3. Click Install
  4. Restart ComfyUI

From Source

Clone/extract the repository into your ComfyUI custom_nodes directory:

cd ~/ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git

Authentication Configuration

Civitai ComfyUI nodes resolve credentials in this strict order:

  1. CIVITAI_API_TOKEN (environment variable)
  2. ~/.civitai/comfy-api-key
  3. ~/.civitai/comfy-oauth.json
  4. CIVITAI_OAUTH_CLIENT_ID

Headless/remote ComfyUI installations should use the environment variable. Add to ~/.bashrc or your ComfyUI startup script:

export CIVITAI_API_TOKEN="your-token-here"

Node Organization

Nodes are organized in the ComfyUI menu in this hierarchy:

Civitai/<media_type>/<ecosystem>[/<engine>]/…

Specific examples:

  • Civitai/Image/ernie › ernie / createImage
  • Civitai/Image/ernie › turbo / createImage
  • Civitai/Image/zImage › zImage / turbo / createImage

Each discriminator variant has its own node, displaying only the inputs that variant actually uses.

Using ERNIE-Image Nodes

Basic ERNIE-Image Generation Node

Add a Civitai/Image/ernie/createImage node to ComfyUI:

  1. Input prompt: Your text description
  2. Set width / height: Recommended 1024×1024
  3. Set steps: Standard mode 20, Turbo mode 8
  4. Set cfgScale: Standard mode 4, Turbo mode 1
  5. Connect to SaveImage node to view output

Output Types

Each node returns:

  • Native Comfy types: IMAGE (images), VIDEO (video), AUDIO (audio)
  • Debug info: workflow_id and raw_json for debugging and cost inspection

LoRA Selector System

Civitai nodes use the AIR URN system to reference models and LoRAs, rather than local file paths.

LoRA Selector Nodes

Use LoRA selector nodes under the Civitai/Loaders category:

  1. Click the 🔍 Browse Civitai button
  2. Search for ERNIE-Image compatible LoRAs
  3. Select the target LoRA
  4. Set strength
  5. Connect to the ERNIE-Image node's loras input

Multi-LoRA Stacking

Each LoRA selector node holds one LoRA. Chain multiple selector nodes via the loras input for stacking.

Key detail: LoRAs take effect through their AIR URN and strength — the generator ignores the keywords field. To actually invoke the trained concept, paste the keywords into the node's prompt.

LoRA Local Download

LoRA selector nodes have a path output. Files are downloaded to local storage only when path is connected. Not connecting path keeps purely cloud mode, pulling no files.

ControlNet Support

ControlNet selector nodes support:

  • Preprocessor selection
  • Weight setting
  • Step range
  • Control image input

Chain multiple ControlNets via the control_nets input.

Browsing History and Import

Civitai Sidebar Tab

The Civitai ComfyUI node pack includes a built-in generation history sidebar:

  • Lists generation history across all media types and sources (web/API/ComfyUI)
  • Filter by media type
  • Click for lightbox preview
  • Scroll to paginate

Import to Canvas

  1. Click thumbnail → Add to canvas: Auto-creates matching loader node
  2. Fill: Populates a selected loader node
  3. Drag: Drop thumbnail directly onto canvas

Development Notes

Nodes Are Auto-Generated

Never manually edit the civitai_comfy_nodes/generated/ directory. To change node shapes:

  • Edit codegen/overrides.json
  • Regenerate

To sync Orchestration API changes:

scripts/sync-spec.sh

Typical Workflow Examples

Workflow 1: Pure Cloud ERNIE-Image Generation

Civitai/ernie/createImage → SaveImage

The simplest cloud generation workflow. No local GPU or model files needed.

Workflow 2: Cloud ERNIE-Image + LoRA

Civitai/LoRA Selector → (loras) → Civitai/ernie/createImage → SaveImage

Use community LoRAs to enhance ERNIE-Image's style or concept.

Workflow 3: Cloud ERNIE-Image + Post-Processing

Civitai/ernie/createImage → ImageScale → SaveImage

Cloud generation followed by local post-processing like scaling.

Cost Optimization Strategies

1. Turbo Mode for Iteration

Turbo mode costs only 40% of standard (8 vs 20 Buzz), ideal for prompt iteration and draft generation.

2. Reasonable Step Count

Diminishing returns for standard mode past 25 steps. Keep Turbo mode at 6-12 steps.

3. Batch Generation

The quantity parameter supports 1-12. Generating multiple images per call is more efficient than multiple calls.

4. Resolution Choice

ERNIE-Image performs best around 1024². Excessively large resolutions increase Buzz cost with limited quality improvement.

MCP Native Integration

Civitai Orchestration API natively supports MCP (Model Context Protocol):

  • 20 tools
  • 3 prompts
  • Blob resources

This means you can connect Claude Desktop, claude.ai, or any MCP-aware client to the same orchestrator, enabling AI-assisted ComfyUI workflow design.

FAQ

Q: What are the advantages over local ComfyUI ERNIE-Image nodes?

Zero hardware cost, no model downloads, automatic updates. The simplest choice for users without high-end GPUs.

Q: How often are nodes updated?

The Civitai team continuously updates nodes, auto-generating new ones from Orchestration API changes. Updates available through ComfyUI Manager.

Q: Which ERNIE-Image variants are supported?

Currently standard and Turbo modes. Additional variants may be added in the future.

Q: How to top up Buzz credits?

Purchase Buzz credits on the Civitai platform. See Civitai official documentation for pricing.

Summary

Civitai ComfyUI custom nodes bring cloud AI capabilities directly into ComfyUI workflows. Key advantages:

  • Zero GPU cost: All computation completed in the cloud
  • 160+ auto-nodes: Covers image, video, and audio generation
  • Native Comfy integration: Returns IMAGE/VIDEO/AUDIO types, seamlessly connecting existing nodes
  • LoRA selectors: Browse, select, and stack community LoRAs
  • AIR URN standardization: Unified model reference standard
  • MCP native: AI-assisted workflow design

For users who want to experience ERNIE-Image without investing in GPU hardware, or creators who need cloud batch generation, Civitai ComfyUI nodes are the best choice.

ERNIE-Image Team