ComfyUI Gains ERNIE‑Image Support – Precise Text Rendering & Structured Image Generation

What the ComfyUI Blog Is Announcing

The latest ComfyUI announcement introduces ERNIE‑Image, Baidu’s open‑source text‑to‑image diffusion model (Apache‑2.0 licence), now available as a built‑in node in ComfyUI. Powered by an 8‑billion‑parameter Diffusion Transformer (DiT) and paired with a 3‑billion‑parameter Prompt Enhancer, the model brings high‑fidelity text rendering, strong instruction following, and a wide stylistic range to the ComfyUI workflow ecosystem.

Key Features in This Release

Precise multilingual text rendering – dense, layout‑sensitive typography works in English, Chinese and other scripts.
Robust instruction following – handles complex prompts, multi‑object relations, and knowledge‑rich descriptions.
Structured visual generation – ideal for posters, manga/storyboards, and multi‑panel compositions.
Broad stylistic spectrum – from realistic photography to cinematic, film‑like aesthetics.
Compact and deployable – 8 B parameters run comfortably on a 24 GB VRAM GPU.
Built‑in Prompt Enhancer – a 3 B model that expands short inputs into richer prompts, improving quality without extra user effort.

Model Variants Available

Variant	Typical Steps	Speed / Quality Trade‑off
ERNIE‑Image (Main SFT)	~50 inference steps	Stronger image quality and instruction fidelity
ERNIE‑Image‑Turbo	8 steps (DMD + RL optimized)	Faster generation for rapid prototyping

Why This Matters for Workflow Creators, Template Authors, and the ComfyUITemplates.com Community

Elevated Text‑Heavy Designs – Precise rendering of titles, subtitles, and annotated diagrams (e.g., the coffee‑making infographic or Japanese flashcard examples) means creators can rely on a single model for both visual content and crisp typography.
Multi‑Panel & Storyboard Support – The model’s structured generation excels at comic‑style layouts, educational panels, and concept posters, giving template authors ready‑made building blocks for “multi‑panel compositions” without stitching separate outputs together.
Cross‑Language Flexibility – Supporting English, Chinese and other scripts opens doors for bilingual templates, expanding the audience for assets listed on ComfyUITemplates.com.
Efficient Resource Use – At 8 B parameters the model fits on common consumer‑grade GPUs, allowing creators to share lightweight, reproducible workflows that can be run locally or on Comfy Cloud.
Prompt Enhancer Integration – Short, natural prompts are automatically enriched, lowering the barrier for less‑experienced users while still giving power users fine‑control over detailed prompts.

How This Affects ComfyUI Templates and Apps

Template Discovery – Users can now search the “ERNIE‑Image” tag in the Template → Search panel, surfacing workflows that leverage the new model.
Template Compatibility – Existing multi‑step pipelines (e.g., combining a layout node with a diffusion node) can replace older diffusion models with ERNIE‑Image without major rewiring, thanks to the familiar node interface.
App Builder & Cloud Deployments – The “Turbo” variant is ideal for real‑time web apps or low‑latency demo galleries, making it easier to showcase ERNIE‑Image‑powered creations on community hubs.
Documentation & Community Support – The official ERNIE‑Image page on Hugging Face supplies model files and usage notes, which template authors can link directly from their workflow descriptions on ComfyUITemplates.com.

Getting Started

Update ComfyUI to v0.19.1 (or launch the latest Comfy Cloud instance).
Open Template → Search and type “ERNIE‑Image”.
Select the desired ERNIE‑Image workflow (Main SFT or Turbo).
Download any missing model weights, customize the prompt, and press Run.

ComfyUITemplates.com

ERNIE-Image Support in ComfyUI: Precise Text Rendering and Structured Image Generation