ERNIE-Image ComfyUI workflow example

ERNIE-Image is Baidu’s open 8B Diffusion Transformer text-to-image model with precise text rendering, strong instruction following, and a built-in Prompt Enhancer.

ERNIE-Image is an open text-to-image model by Baidu, licensed under Apache-2.0. Built on an 8B parameter Diffusion Transformer (DiT), it delivers high-quality image generation with precise text rendering, strong instruction following, and structured visual generation.

The model includes a built-in Prompt Enhancer (3B) that expands short inputs into richer prompts for better results.

Model highlights:

• Precise text rendering — dense and layout-sensitive text in English, Chinese, and more
• Strong instruction following — handles complex prompts, multi-object relations, and knowledge-intensive descriptions
• Structured visual generation — posters, manga/anime storyboards, multi-panel compositions
• Broad stylistic range — realistic photography to cinematic film-like aesthetics
• Compact and deployable — 8B parameters, runs on 24 GB VRAM
• Built-in Prompt Enhancer — 3B model that expands short inputs into richer prompts

Related links:

• Hugging Face (ERNIE-Image)
• Hugging Face (ComfyUI Support)

ERNIE-Image text-to-image workflow

Download Workflow

Download the ERNIE-Image text-to-image workflow JSON file.

Run on Comfy Cloud

Run this workflow directly on Comfy Cloud.

Portable or self deployed users

Make sure your ComfyUI is updated.

• Download ComfyUI
• Update Guide

Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated. (Desktop version’s update will delay sometime)

If nodes are missing when loading a workflow, possible reasons:

1. You are not using the latest ComfyUI version (Nightly version)
2. Some nodes failed to import at startup

Desktop or Cloud users

• The Desktop is base on ComfyUI stable release, it will auto-update when there is a new Desktop stable release available.
• Cloud will update after ComfyUI stable release.

So, if you find any core node missing in this document, it might be because the new core nodes have not yet been released in the latest stable version. Please wait for the next stable release.

Get started

1. Update ComfyUI to the latest version or use Comfy Cloud
2. Go to Template and search for ERNIE-Image
3. Select the ERNIE-Image workflow
4. Download any missing models, update the prompt, and click Run

ERNIE-Image model downloads

You can find all repackaged model files at Comfy-Org/ERNIE-Image on Hugging Face.

ernie-image.safetensors

Diffusion model for ERNIE-Image.

ministral-3-3b.safetensors

Text encoder for ERNIE-Image.

ernie-image-prompt-enhancer.safetensors

Prompt Enhancer text encoder for ERNIE-Image.

flux2-vae.safetensors

VAE for ERNIE-Image.

Model storage location

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── ernie-image.safetensors
│   ├── 📂 text_encoders/
│   │   ├── ministral-3-3b.safetensors
│   │   └── ernie-image-prompt-enhancer.safetensors
│   └── 📂 vae/
│       └── flux2-vae.safetensors

ERNIE-Image-Turbo

ERNIE-Image-Turbo is a faster variant optimized with DMD and RL, generating images in just 8 steps compared to the ~50 steps required by the standard model.

Download Workflow

Download the ERNIE-Image-Turbo text-to-image workflow JSON file.

Run on Comfy Cloud

Run this workflow directly on Comfy Cloud.

ERNIE-Image-Turbo model downloads

ernie-image-turbo.safetensors

Diffusion model for ERNIE-Image-Turbo.

ministral-3-3b.safetensors

Text encoder for ERNIE-Image-Turbo.

ernie-image-prompt-enhancer.safetensors

Prompt Enhancer text encoder for ERNIE-Image-Turbo.

flux2-vae.safetensors

VAE for ERNIE-Image-Turbo.

Model storage location

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── ernie-image-turbo.safetensors
│   ├── 📂 text_encoders/
│   │   ├── ministral-3-3b.safetensors
│   │   └── ernie-image-prompt-enhancer.safetensors
│   └── 📂 vae/
│       └── flux2-vae.safetensors

Available models

Model	Description	Inference steps	Link
ERNIE-Image	Main SFT model — stronger quality and instruction fidelity	~50	Hugging Face
ERNIE-Image-Turbo	Turbo model optimized with DMD and RL — faster generation	8	Hugging Face

Examples

Text rendering and design layouts

Prompt

Educational comic book style infographic showing the coffee making process. The background has a light beige vintage paper texture. At the top center of the image is a bold brown title that reads 'The Coffee Making Process', with a smaller English subtitle 'How Coffee is Made' below. The main part consists of six step blocks connected by brown dotted arrows, arranged in two rows up and down, forming a Z-shaped visual guide.

The first step in the upper left corner illustrates a coffee tree full of ripe red coffee cherries, with a cut coffee cherry next to it showing the beans inside, labeled 'Step 1: Harvesting Cherries' below the block.
The second step in the upper middle shows wooden fermentation boxes filled with coffee beans, labeled 'Step 2: Pulping and Fermentation'.
The third step in the upper right depicts coffee beans being dried under the sun on bamboo mats, labeled 'Step 3: Sun Drying'.
The fourth step in the lower left shows a vintage metal roasting machine with coffee beans rolling inside and steam rising, labeled 'Step 4: Roasting'.
The fifth step in the lower middle features a stone grinder pouring out smooth coffee grounds, labeled 'Step 5: Grinding'.
The sixth step in the lower right shows a production line where coffee liquid is being poured into molds, with finished packaged coffee beside it, labeled 'Step 6: Brewing and Forming'.

The four corners of the image are decorated with hand-drawn coffee leaves and coffee beans. The overall color palette consists of warm brown, caramel, cream, deep red and olive green, with delicate lines and clear layout.

Prompt

Vertical rectangular language learning flashcard in traditional Japanese Ukiyo-e woodblock style. Whole card framed by dark blue traditional Seigaiha wave borders, with aged beige rough washi paper texture background.

Upper two-thirds: Ukiyo-e illustration. A graceful spotted sika deer stands calmly in the center, with delicate black outlines and traditional flat coloring. Background: classic Japanese scenery — snow-capped Mount Fuji in distance, stylized layered clouds, falling pink cherry blossom petals. A red square antique seal with the Chinese character '鹿' in seal script at the lower right. Color scheme: indigo, vermilion, moss green, beige, with woodblock color misalignment and vintage charm.

Lower one-third: centered text area. Top: small hiragana しか; large Japanese kanji 鹿; below it romaji shika; then English "Deer". Bottom: bilingual example — Japanese: 鹿が桜の花を見ている。English: The deer is looking at the cherry blossoms. All black text with slight brush calligraphy texture, matching Ukiyo-e antique style.

Prompt

Split-screen conceptual poster with vertical split composition, left half representing the present and future featuring a solid stone analog clock dial with bright red hands and red scale markings, set against a deeply cracked, dry, parched earth texture in muted cool gray tones, embodying slow, static structural decay and temporal erosion, right half representing the past that is violently disintegrating, dissolving and exploding into chaotic dust, debris, flying stone shards and swirling cosmic nebula energy with a deep blood red and black color palette, the circular clock frame itself is half solid weathered stone and half crumbling into particles merging the two worlds, with the red clock hands fully positioned on the left present/future side of the dial, embodying the concept of time, past vs future, slow decay vs violent collapse, hyper-realistic 3D render with cinematic dramatic lighting, high contrast, sharp details on the left side, motion blur and dynamic particle effects on the right side, photorealistic textures of cracked earth and floating dust, atmospheric haze, futuristic graphic design with minimalist red vertical UI elements and custom technical text overlays in the corners, bold red stylized title text "ETERNAL DAWN" in the top right corner, small red technical metadata text "SPECDARY BOTHUNG" directly below the main title in the top right, dense lines of small red technical data blocks including "TIMESTAMP: 00:00:00", "COLLAPSE RATE: 99.8%", "TEMPORAL ANOMALY DETECTED", "PAST DIMENSION: DECAYING" in the bottom right corner with a red horizontal accent bar and footer text "FOR DECAY OF THE EONS" below, small red header text "TEMPORAL SHIFT PROTOCOL" with vertical red accent line running down the left edge in the top left, vertical red accent line with small red technical text annotations including "PRESENT: STABLE", "FUTURE: UNFOLDING" along it in the bottom left, red text "WARNING: TEMPORAL DISINTEGRATION" curving around the center clock dial perimeter, small red footer text "PAST FADES, FUTURE RISES" spanning the split at the bottom center, shot in the style of a high-end sci-fi movie poster, 8K ultra-high resolution, photorealistic, cinematic composition, dramatic depth, moody and intense atmosphere, detailed particle simulation, photorealistic material rendering.

Cinematic and stylized aesthetics

Prompt

First-person vertical real-world urban night photo. Narrow, wet, busy city street with strong depth; towering buildings on both sides with fire escapes, pipes, AC units, warm window lights. Red vertical "HOTEL" neon on left, bright blue "BAR" sign and yellow "OPEN 24 HOURS" box on right. Distant huge digital billboard glowing cyan "NEON CITY". Rough wet asphalt reflects neon lights; vintage yellow taxi with "TAXI" sign drives down center, red taillights on. Two pedestrians in dark coats with black umbrellas walk away on right sidewalk. Giant needle-like steel TV tower peeks through building gaps at end, red aircraft warning light glowing. Cinematic volumetric light, misty air, cool cyan-blue vs warm orange-red contrast, immersive and realistic.

Prompt

Full-body editorial fashion shot, elegant European woman aged 28-30 with soft natural blonde hair styled in a loose low bun, wearing a tailored soft cream ribbed knit long-sleeve maxi dress with a fitted waist and flowing floor-length skirt, minimalist nude leather mules and delicate gold jewelry, standing leaning against an ornate carved stone balustrade on a gravel terrace in the garden of a grand Lake Como villa, tall iconic Italian cypress trees framing the background, calm still Lake Como water stretching out to hazy misty Alpine mountains in the distance, overcast soft diffused natural light, muted warm earthy color palette, quiet luxury aesthetic, hyper-realistic editorial fashion photography, natural skin texture with subtle pores, soft bokeh foreground foliage, shallow depth of field, shot on a medium format camera with an 85mm f/1.4 lens, sharp focus on the subject, fine film grain, muted contrast, timeless sophisticated mood, shot by Lachlan Bailey in the style of Vogue editorial, 8K ultra-high resolution, photorealistic, cinematic composition following the rule of thirds, golden hour overcast lighting, atmospheric haze, refined Italian villa garden details.

Prompt

A charming, whimsical 3D illustration featuring a fluffy, plush-textured white duck sitting comfortably in a fuzzy coral-pink armchair, holding a bright red mug of steaming black coffee against a solid warm coral-red background; defined by its tactile, huggable fabric-like texture across all elements, a bold, minimalist warm color palette of creamy white, vibrant orange, and rich reds, gentle anthropomorphism that blends cuteness with relatable cozy relaxation, soft diffused lighting, and a subtle film grain that adds a nostalgic, handcrafted feel, creating a playful yet serene mood perfect for themes of calm downtime and morning routines.

Multi-panel compositions

Prompt

Educational comic book infographic. Five vertical panels side by side, earthy color palette, hand-drawn illustration style. Top title: "NORTH AMERICAN NATIVE SPECIES". Style: comic line art, watercolor fills, white panel backgrounds, bold headers.
Panel 1: Gray squirrel holding an acorn. Header: "EASTERN GRAY SQUIRREL" Fact: "Buries acorns to help forests grow." Arrow → tail: "Bushy tail"

Panel 2: Robin with orange-red breast on a branch. Header: "AMERICAN ROBIN" Fact: "A sign of spring returning." Arrow → breast: "Orange-red breast"

Panel 3: Red maple tree with autumn leaves. Header: "RED MAPLE TREE" Fact: "Shelters wildlife year-round." Arrow → leaves: "Red in autumn"

Panel 4: Deer with white tail raised. Header: "WHITE-TAILED DEER" Fact: "White tail signals danger." Arrow → tail: "Alarm signal"

Panel 5: Monarch butterfly on a flower. Header: "MONARCH BUTTERFLY" Fact: "Migrates 3,000 miles each year." Arrow → wings: "Warning colors"

Prompt

A 6-panel comic page, 2 columns × 3 rows, black border between each panel.

Panel 1 (top-left): wide shot — a young woman in a red coat stands at a rainy train station. Caption box: "She had waited three years for this moment."
Panel 2 (top-right): close-up on her face — anxious eyes, rain on her cheek. No text.
Panel 3 (mid-left): medium shot — a train arrives, doors slide open, steam rising. Sound effect text: "WHOOOOSH"
Panel 4 (mid-right): her point-of-view — a man in a gray jacket steps out, back to camera.
Panel 5 (bottom-left): extreme close-up — her hand trembling as she reaches forward.
Panel 6 (bottom-right): wide shot — they face each other under one umbrella. Caption box: "Some arrivals change everything."

Ink and watercolor style, cool blue-gray palette, expressive line art. Same character design consistent across all 6 panels.