Run HiDream in ComfyUI: Easy Step-by-Step Guide

The world of AI-generated art is evolving fast, and one of the most exciting developments is the native integration of HiDream into ComfyUI. This marks a major milestone for creators who previously relied on custom NF4 wrapper nodes to access HiDream’s capabilities. Now, with native support, the setup is faster, more intuitive, and far more accessible—even for those with limited hardware. Users can benefit from enhanced functionality like improved text encoding via four specialized encoders, standard KSampler support, and new quantized GGUF versions that perform well even on systems with less than 16GB of VRAM.

Whether you're aiming for high-fidelity generations or efficient rendering on cloud GPUs, this update opens the door to a smoother, more powerful creative workflow. In this guide, we’ll walk you through everything you need to get started—from installation to performance comparisons—so you can take full advantage of HiDream’s features in ComfyUI.

HiDream offers a flexible set of model variants tailored to different hardware capacities and creative needs. Whether you're aiming for maximum quality, a balance of performance and efficiency, or lightning-fast generation for prototyping, there's a HiDream version designed for your workflow. Below, we break down the key differences and help you choose the right model for your setup.

HiDream-I1 Model Variants (Official)

HiDream-AI provides three primary model versions, each optimized for different use cases:

• HiDream-I1-Full: The highest quality output with the most detailed renderings. Requires the most VRAM (~20GB).

• HiDream-I1-Dev: A distilled model with fewer steps but excellent balance between quality and speed. Recommended for most users (needs ~16GB VRAM).

• HiDream-I1-Fast: Prioritizes speed with just 16 steps. Ideal for quick drafts or limited VRAM setups (~12GB or less).

🌍 Community Variants & Optimized Versions

Several community-backed and hardware-friendly versions of HiDream exist to support a wider range of users:

GGUF Quantized Models

Provided by [city96], these use the new GGUF format to reduce file size and memory usage without fully sacrificing quality.

✅ Requires the ComfyUI-GGUF extension and the Unet Loader (GGUF) node.
📦 These models are available in Q8, Q6, Q5, Q4, Q3, Q2 quantization levels.

NF4 Quantized Models

These use 4-bit quantization to lower memory requirements and run efficiently on ~16GB VRAM setups.

• HiDream-I1-Full-nf4

• HiDream-I1-Dev-nf4

• HiDream-I1-Fast-nf4

Requires the ComfyUI-HiDream-Sampler (by lum3on). This special node:

• Downloads models automatically on first use

• Adds unofficial image-to-image support

Repackaged HiDream Versions (ComfyOrg)

ComfyOrg maintains ready-to-use versions of HiDream models, including:

• Full versions

• FP8-quantized models (~16GB VRAM usage)

These work seamlessly with native ComfyUI nodes and require no external scripts or wrappers.

⚙️ Summary: Which Model Should You Use?

With a clear understanding of the available HiDream model versions—ranging from Full to Dev to Fast, and their various formats like FP8, GGUF, and NF4—you’re ready to dive into the practical setup.

In the next section, we’ll start by showing you how to run the HiDream-I1-Dev FP8 version, which offers a great balance between quality and performance with minimal setup. Once that’s running, we’ll introduce an alternative GGUF version for users with lower VRAM or different performance needs, and later compare the two in detail to help you choose what’s best for your workflow.

Before diving into HiDream’s powerful capabilities in ComfyUI, you’ll need to set up your environment correctly. Whether you're running locally or on a cloud GPU platform like RunPod, this section will guide you through everything you need to get started with the HiDream-I1-Dev FP8 version—our recommended starting point for its excellent balance of quality, performance, and setup simplicity.

1. Install ComfyUI

HiDream runs inside ComfyUI, a flexible and powerful node-based interface for AI image generation. If you haven't installed it yet, choose the setup that fits your needs:

• Local Installation
  Ideal if you have a powerful PC or workstation (12GB+ VRAM recommended).
  👉 Follow our Local ComfyUI Setup Guide.

• RunPod Deployment
  Best for users who prefer cloud GPU access with minimal setup.
  👉 How to run ComfyUI on RunPod.

2. Download the Required Files

To run HiDream in ComfyUI, you’ll need to manually download a few essential components. These are not bundled by default with ComfyUI.

Four CLIP Text Encoders

HiDream uses four separate CLIP encoders to improve semantic understanding. These must be placed in the correct directory.

• 🔗 clip_l_hidream.safetensors

• 🔗 clip_g_hidream.safetensors

• 🔗 t5xxl_fp8_e4m3fn_scaled.safetensors

• 🔗 llama_3.1_8b_instruct_fp8_scaled.safetensors

• 📁 Save to: ComfyUI/models/text_encoders/

Variational Autoencoder (VAE)

HiDream requires a specific VAE (same as Flux VAE) for decoding latent representations into final images.

• 🔗 ae.safetensors

• 📁 Save to: ComfyUI/models/vae/

HiDream Model Checkpoint

There are multiple versions of HiDream. We recommend starting with the hidream_i1_dev_fp8.safetensors for optimal balance.

• 🔗 hidream_i1_dev_fp8.safetensors

• 📁 Save to: ComfyUI/models/diffusion_models/

(You can swap this later with a GGUF model in the upcoming section.)

3. Verify Your Setup

After downloading and placing everything in the correct folders:

ts
1📂 ComfyUI/
2├── 📂 models/
3│   ├── 📂 text_encoders/
4│   │   ├─── clip_l_hidream.safetensors
5│   │   ├─── clip_g_hidream.safetensors
6│   │   ├─── t5xxl_fp8_e4m3fn_scaled.safetensors
7│   │   └─── llama_3.1_8b_instruct_fp8_scaled.safetensors
8│   └── 📂 vae/
9│   │   └── ae.safetensors
10│   └── 📂 diffusion_models/
11│       └── hidream_i1_dev_fp8.safetensors  # Or any other HiDream (GGUF) model you choose           

With ComfyUI installed and the Dev FP8 model files in place, you're ready to start building your workflow. In the next section, we’ll show you how to load and configure the HiDream workflow in ComfyUI using this model—step-by-step.

Now that your environment and model files are in place, it’s time to load and configure the HiDream-I1-Dev FP8 workflow in ComfyUI. This setup ensures that all components work together correctly for optimal image quality, stability, and performance. Once configured, you'll be ready to begin testing HiDream using real prompts.

1. Load the HiDream Workflow

Start by loading the HiDream-I1-Dev FP8 workflow:

👉 Download the provided HiDream workflow JSON file and drag it into your ComfyUI canvas.

This template includes all required nodes for HiDream, pre-arranged for easy customization and testing.

2. Configure the Nodes

Double-check each node in the workflow to ensure it references the correct model files and settings:

3. Test with a Showcase Prompt

With your workflow now fully configured, let’s put HiDream Dev FP8 to the test using a high-impact example prompt. This helps you instantly evaluate its artistic quality, rendering detail, and how well the model handles style and sensuality.

This render took around ~20seconds on a RTX 4090 (24GB VRAM). Keep in mind, the first render might take a bit longer as the system initializes.

✨ Example Prompt:

A divine, hyperrealistic woman is seated in a relaxed pose against a black void, framed from upper body down. Her translucent silk outfit reveals her curves and prominent cleavage, flowing like vapor around her form. Her white-grey hair falls behind her shoulders in a ponytail, catching subtle cinematic light. Her blue eyes are fixed on the viewer, warm and lifelike. White stockings add sensual detail to her fully visible legs. The scene is softly dreamlike, with swirling fog and barely-there angel wings glowing in the background. Photorealistic lighting and skin texture.

Ready to unlock reduced VRAM usage? The HiDream Dev GGUF models come in various quantization levels designed to fit different GPU memory sizes — from highly compressed Q2 versions up to full precision BF16 models.

👉 Step 1: Choose the Right Model for Your VRAM

As a rule of thumb, pick the highest-quality GGUF model that fits comfortably in your GPU’s VRAM. For example, if you have 16GB of VRAM, the Q6 model (around 14.7GB) is a perfect choice, balancing image fidelity and memory efficiency.

👉 Step 2: Install the ComfyUI-GGUF Extension & Load the Workflow

To use GGUF models, you need the ComfyUI-GGUF extension. Here’s how to get it:

• Open ComfyUI and go to Manager → Custom Node Manager

• Search for ComfyUI-GGUF (developed by user city96)

• Install the node and then restart ComfyUI

• After that, download the dedicated HiDream Dev GGUF workflow JSON file and drag it into your ComfyUI canvas.

  

• Make sure to select your downloaded GGUF model file inside the Unet Loader (GGUF) node to properly load the model.

👉 Step 3: Test and Compare

Run your favorite prompts using the GGUF setup to compare rendering speed, detail, and overall quality with the FP8 version. This hands-on test will give you a clear sense of which model suits your creative workflow best.

This render took approximately ~135 seconds on an RTX 2000 Ada (16GB VRAM). Keep in mind, the first render might take a bit longer as the system initializes.

✨ Example Prompt:

In a dreamy, hyperrealistic scene, a woman sits on the right side of the canvas, framed close-up from the waist up. Her translucent silk outfit flows over her curves, accentuating her cleavage and midriff, glowing faintly in the soft light. Her white-grey hair falls gracefully behind her shoulders, with a subtle backlight catching the strands. Her bright blue eyes gaze directly into the camera, inviting and lifelike. The surrounding fog gently envelops her, while barely-there angel wings glow faintly behind, completing the serene and divine mood. Photorealistic lighting emphasizes the soft texture of her skin and attire.

Now that both the HiDream Dev FP8 and HiDream Dev GGUF (Q5) setups are tested and ready, it’s time to dive deeper. We'll run a curated series of prompts designed to push the HiDream model to its creative and technical limits—comparing both variants along the way. This hands-on evaluation will help reveal how well HiDream handles anatomy, lighting, detail, composition, and more—and whether there are any noticeable differences between the FP8 and GGUF experiences.

To truly push the limits of what the HiDream Dev FP8 and HiDream Dev GGUF (Q5) models can do, we crafted a series of demanding prompts designed to stretch their capabilities across both technical precision and artistic flair. These tests zero in on everything that makes or breaks a high-quality AI render—like anatomical accuracy, lighting mastery, text clarity, stylistic control, and sensual detail handling.

By comparing both versions side-by-side, we’re not just looking at raw output—we’re evaluating HiDream as a creative tool. The goal? To see how well it performs when challenged to produce images that are not only technically clean but also visually captivating, emotionally expressive, and styled with intentional sensuality.

Now, let’s break it down—category by category—and see where each model shines. Let's start with the first one.

1. Anatomy & Proportions

• Human anatomy:

  
  Prompt: “A glamorous portrait of a confident woman looking directly at the camera, wearing a sexy leather outfit with visible cleavage. Her perfectly proportioned facial features and detailed hands delicately touch her exposed collarbone. Softly glowing skin under warm, flattering studio lighting with a clean, elegant background.”

• Animal anatomy:

  
  Prompt: “A fierce tigress mid-leap through a lush jungle, muscles rippling realistically and eyes focused intensely, captured from a dynamic low-angle shot highlighting her powerful movement.”

• Complex poses:

  
  Prompt: “A seductive model gracefully performing the Tree Pose (Vrksasana) on a minimalist set, balancing on one leg with her hands raised above her head in a gentle arch that accentuates her natural curves and cleavage. She gazes directly into the camera, wearing a sleek, form-fitting bodysuit that highlights flawless anatomy. Soft, flattering lighting reveals the smooth contours of her body and the strength in her pose.”

2. Detail Handling

• Micro details:

  
  Prompt: “Extreme close-up of a dewdrop resting on a rose petal, light sparkling through with crystal clarity. Nearby, a fingertip shows realistic skin pores and fine hairs, highlighting exquisite texture precision.”

• Macro scenes:

  
  Prompt: “Lavish twilight penthouse party filled with elegantly dressed guests. Fabrics like silk, velvet, and leather showcase intricate textures, while sparkling jewelry and ambient lighting create a glamorous atmosphere.”

• Textures:

  
  Prompt: “Extreme close-up of premium materials—rich velvet upholstery with soft folds, smooth satin ribbons tied around a sleek perfume bottle, droplets of condensation on polished glass, and intricate lace textures overlaying elegant packaging."

  Prompt: “Macro shot of a Rolex watch face with a rich green dial, gold-finished bezel and hands, highlighting the polished metal textures, glossy sapphire crystal, and intricate engravings.”

3. Text Rendering

• Simple text:

  
  Prompt: “A glowing neon sign reading ‘Enter the Pleasure Lounge’ in a refined, flowing cursive font. The cool white and deep blue illumination softly reflects on a matte dark wall, creating a modern, upscale atmosphere.”

4. Style Versatility

• Art Deco Glamour:

  
  Prompt: “A sultry portrait inspired by 1920s Art Deco style, featuring sharp geometric shapes and bold lines that emphasize glamorous curves and pronounced cleavage. The composition uses rich gold and black tones with elegant, cinematic lighting to evoke vintage luxury and sophistication.”

• Charcoal Sketch Realism:

  
  Prompt: “A sensual figure rendered in a realistic charcoal sketch style, with soft, smudged shadows and delicate highlights enhancing glowing skin textures. The expressive linework captures raw tactile quality, natural curves, and intimate sensuality.”

• Retro Pinup Pop Art:

  
  Prompt: “A vibrant 1980s-inspired retro pinup scene in bold Pop Art style, bursting with neon colors, graphic outlines, and confident poses. The styling highlights voluminous hair, shimmering fabrics, and dramatic lighting that accentuate feminine strength and playful allure.”

• Mid-Century Modern Cinematic:

  
  Prompt: “A seductive woman styled in mid-century modern aesthetic, leaning against a glowing jukebox in a 1950s diner. Neon reflections and moody shadows create a cinematic atmosphere, blending vintage Americana with sensual confidence and sharp leather textures.”

• Impressionist Nature:

  
  Prompt: “A sleek, futuristic lounge surrounded by towering, bioluminescent tropical leaves with glowing veins in electric blue and neon green. The scene is bathed in soft, ambient light with reflections on glossy metallic surfaces. Wisps of vapor curl through the air, adding mystery and sensuality. In the background, holographic palm fronds sway gently, blending nature with advanced technology. The overall mood is seductive and immersive, fusing tropical lushness with high-tech elegance.”

5. Composition & Concepts

• Abstract concepts:

  
  Prompt: “An artistic visual representing desire and empowerment, featuring flowing, dynamic lines and warm, inviting colors that gracefully highlight the female form. The composition balances softness and strength, evoking emotion through minimalist yet expressive shapes.”

• Complex scenes:

  
  Prompt: “A luxurious boudoir scene with a woman reclining elegantly on silk sheets, softly illuminated by flickering candlelight. The room is filled with opulent décor—velvet cushions, ornate mirrors, and subtle gold accents—creating an intimate and sensuous atmosphere.”

• Storytelling:

  
  Prompt: “A confident woman in a glossy white swimsuit edged with gold chains lounges on the sun-drenched deck of a private yacht. Cat ears with crystal embellishments glint in the sunlight as she reclines against plush cushions, one hand shading her eyes from the sparkling blue ocean. The scene captures sleek, playful elegance amid summer luxury.”

  

  Prompt: "A playful woman in a vibrant red bikini with gold chains lounges poolside on an oversized float shaped like a flamingo. Cat ears with jeweled accents peek through her wet, tousled hair as she blows a kiss toward the camera. The crystal-clear water sparkles under the midday sun, evoking fun and flirtatious luxury."

  

  Prompt: "A fierce woman in a daring leather harness and flowing sheer cape stands atop a sunbaked dune at golden hour. Her bold makeup contrasts with the natural landscape, and sleek cat-eye sunglasses add an edge. The wind sweeps her cape dramatically, capturing raw power and sensual mystique."

6. Technical Challenges

• Lighting conditions:

  
  Prompt: “A dramatic chiaroscuro portrait capturing a woman half bathed in soft, warm golden light while the other half fades into deep shadow. The interplay of light and darkness accentuates her curves and cleavage, creating a moody, cinematic effect with rich contrast and subtle skin texture.”

• Reflections and transparency:

  
  Prompt: “A crystal-clear glass of red wine held delicately in manicured hands, with intricate reflections bouncing off polished nails and smooth, glowing skin. The scene showcases realistic transparency, light refraction, and subtle highlights to test reflective detail.”

• Perspective:

  
  Prompt: “A daring upward perspective focusing on a woman perched on the edge of a sleek glass table. She arches her back elegantly, wearing a tight leather corset with visible strappy details wrapping around her waist and thighs. Her stiletto heels rest firmly on the floor, legs crossed at the ankles to create dynamic lines. The transparent glass surface reveals her silhouette from below, accentuating the tension in her posture and the smoothness of her exposed skin. Dramatic side lighting creates striking highlights and shadows, enhancing the glossy textures and sultry vibe.”

  

  Prompt: "A captivating fisheye lens perspective from slightly below, showing a confident woman reclining on a plush velvet sofa. She wears a strappy leather harness and thigh-high boots, legs extended toward the camera and slightly parted, revealing intricate lace details beneath. The curvature of the lens accentuates her curves and the glossy textures of her outfit, while warm ambient lighting highlights her glowing skin and deep cleavage. The soft-focus background creates intimacy, making her the undeniable focal point of this bold, sensual composition."

  

  Prompt: "From above, a woman kneels on a luxurious fur rug, dressed in a sheer black mesh bodysuit with strategically placed opaque panels, teasingly revealing cleavage and curves. Her hair is styled in soft pigtails, framing her flushed face. Her mouth is wide open, tongue sticking out boldly, creating a provocative, sultry expression. Warm, ambient lighting softens the scene, enhancing the texture of the fur and the gleam of her skin.

Across all examples, the model demonstrates impressive command over challenging visual elements—handling lighting, reflections, and perspective with striking accuracy and stylistic coherence. Each prompt delivers seductive, cinematic results with remarkable visual fidelity. While minor anatomical quirks may surface in edge cases or unconventional aspect ratios, the overall output remains consistently refined and production-ready. Whether you're exploring creative concepts or delivering polished work, HiDream proves itself as a powerful, expressive model—now seamlessly integrated into our own custom ComfyUI setup. Get started with HiDream and experience fast renders, stunning quality, and total control, right here on Next Diffusion.

Understanding the trade-offs between different HiDream model versions helps you optimize your setup:

Summary: Use the FP8 models when image quality is paramount and you have sufficient VRAM. Choose GGUF versions for better VRAM optimalisation, especially on consumer-grade GPUs or cloud setups with limited resources.

The native integration of HiDream into ComfyUI isn’t just a technical upgrade—it’s a creative breakthrough. No more clunky workarounds or complex wrappers. Now, creators of all levels can harness HiDream’s full potential—semantic control, dynamic negative prompting, and advanced model performance—all inside a sleek, intuitive workflow. Whether you're crafting seductive visuals, testing technical precision, or pushing artistic boundaries, this setup empowers rapid iteration with pro-level results. And with constant evolution from the community and beyond, the future of AI-powered art has never looked this smooth. Ready to see what’s possible? Dive into HiDream and start creating.