HiDream-E1-1: ComfyUI Setup for Image Editing on Runpod

Welcome to this comprehensive tutorial on the HiDream image editing model (HiDream-E1-1), tailored for use with ComfyUI on RunPod. This powerful tool enables users to edit existing images through prompts, making it ideal for both personal creativity and commercial projects. Released under the MIT license, HiDream offers you the freedom to use it without restrictions—even commercially.

Before jumping into installation and setup, it’s important to understand the requirements and steps needed to ensure a smooth and successful experience.

In this guide, we’ll walk you through all the prerequisites, file downloads, and configurations necessary to get HiDream-E1-1 up and running efficiently. Plus, we’ll explore practical use cases to help you unlock the full potential of this versatile image editing model.

Let’s get started!🎉

To run the HiDream-E1-1 image editing model smoothly, you’ll need a GPU with at least 24GB of VRAM (e.g., NVIDIA RTX 4090). Lower-spec GPUs may cause slowdowns or runtime errors. That’s why we recommend using RunPod, which offers cost-effective access to high-end GPUs—perfect for this workload.

Step 1: Fund Your RunPod Account and Unlock a Bonus

• Create an account and Deposit your first 10$ on RunPod and get a chance to receive up to 500 dollar bonus credits!

Step 2: RunPod Pod Setup: GPU & Storage Configuration

- In the left sidebar, click on the Pods tab.

• At the top of the Pods page, configure the following settings:

  • Set the compute type to GPU.

  • Choose Secure Cloud as your environment.

  • Under Network Volume, select your existing volume. In our case, this is a pre-created volume named ComfyUI Storage (150GB).
    – If you haven't set one up yet, click "Add a Network Volume" and create a new volume with at least 100GB of storage.

  • Next, select the GPU instance you want to use.

  • Choose a GPU with at least 24GB of VRAM — for example, the RTX 4090 (24GB VRAM).

Using persistent network volume like this ensures faster pod reboots and avoids re-downloading models each time you restart.

Want a step-by-step guide on setting up ComfyUI with RunPod and a Network Volume?

💡 Check out this tutorial: How to Run ComfyUI on RunPod

Step 3: Deploy Your GPU Instance

- Scroll down on the pod setup page where you previously selected your GPU (e.g., RTX 4090).

• Look for the template named Next Diffusion - ComfyUI. If you can’t find it, click the “Change Template” button, then search for and select "Next Diffusion - ComfyUI" from the list.

• Once selected, scroll to the bottom and click “Deploy On-Demand” to start your pod.

Step 4: Access Your Pod Workspace

- Once the pod is deployed and running, open your pod’s dashboard.

• Click the Connect button.

• Select HTTP Service → :8888 to open the VS Code workspace in your browser.

  

Step 5: Batch Download All Required Files

After clicking on the HTTP Service port :8888, we’re ready to download the necessary files. The table below provides an overview of what will be downloaded. Don’t worry—these won’t be downloaded manually one by one. Instead, we’ll batch-download them directly within the pod instance using the code block provided below. Let’s read along.

Simply open your VS Code workspace interface via HTTP Service → :8888, then follow the steps below to download all necessary models in one go:

1. In the VS Code interface, click "Toggle Panel" (Ctrl-J) at the top right corner to open the integrated terminal.

2. Paste the command block provided below into the terminal and press Enter to run the script and download all required models in batch.

   ts
   1
   2cd /workspace
   3
   4wget -O ComfyUI/models/diffusion_models/hidream_e1_1_bf16.safetensors https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/resolve/main/split_files/diffusion_models/hidream_e1_1_bf16.safetensors
   5
   6wget -O ComfyUI/models/clip/clip_g_hidream.safetensors https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/resolve/main/split_files/text_encoders/clip_g_hidream.safetensors
   7wget -O ComfyUI/models/clip/clip_l_hidream.safetensors https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/resolve/main/split_files/text_encoders/clip_l_hidream.safetensors
   8wget -O ComfyUI/models/clip/llama_3.1_8b_instruct_fp8_scaled.safetensors https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/resolve/main/split_files/text_encoders/llama_3.1_8b_instruct_fp8_scaled.safetensors
   9wget -O ComfyUI/models/clip/t5xxl_fp8_e4m3fn_scaled.safetensors https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/resolve/main/split_files/text_encoders/t5xxl_fp8_e4m3fn_scaled.safetensors
   10
   11wget -O ComfyUI/models/vae/ae.safetensors https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/resolve/main/split_files/vae/ae.safetensors
   12

   💡 Note: This script automatically downloads all necessary files into the correct folders within your workspace directory. There’s no need for manual downloads or folder creation. Please be patient, as downloading these large files may take some time.

Step 6: Verify Folder Structure

Once the downloads are complete, your pod instance—accessible via port :8888—should reflect the following folder structure within the VS Code workspace:

ts
1📁 ComfyUI/
2└── 📁 models/
3    ├── 📁 clip/
4    │   ├── clip_g_hidream.safetensors
5    │   ├── clip_l_hidream.safetensors
6    │   ├── llama_3.1_8b_instruct_fp8_scaled.safetensors
7    │   └── t5xxl_fp8_e4m3fn_scaled.safetensors
8    ├── 📁 diffusion_models/
9    │   └── hidream_e1_1_bf16.safetensors
10    └── 📁 vae/
11        └── ae.safetensors

Step 7: Launch ComfyUI from the Workspace

With everything set up, you're now ready to launch ComfyUI from your pod instance. Navigate to the root directory (workspace), then paste the following command into the terminal and press Enter to execute the launch script:

ts
1./run_gpu.sh

Once the script is running, wait until the ComfyUI template is fully loaded. You’ll know it’s ready when you see a message in the terminal similar to:

At this point, you have two options to access the ComfyUI interface:

1. Click the link in the terminal by holding 'CTRL' and click the link (http://0.0.0.0:8188) to open the GUI directly in your browser or

2. Access it through the RunPod dashboard:
   Navigate to the Pods section from the left sidebar, locate your running pod settings, click “Connect”, and then select the HTTP Service on port :8188 to launch the GUI.

With ComfyUI running, you're now ready to load the HiDream-E1-1 workflow into the interface.

Step 1: Download the Workflow File

Download the workflow file using the link below:

👉 Download HiDream-E1-1 Workflow JSON

Step 2: Upload the Workflow into ComfyUI

1. Open the ComfyUI interface in your browser (port :8188).

2. Drag and drop the downloaded .json file directly onto the canvas area.

3. The workflow graph will appear automatically.

Step 3: Install Missing Custom Nodes

After loading the workflow, you may notice some nodes highlighted in red—this indicates that required custom nodes are not installed yet.

To resolve this:

1. Click the “Manager” button located in the top-right corner of the ComfyUI interface.

2. In the menu that appears, click “Install Missing Custom Nodes”.

3. A list of missing node packages will be displayed.

4. Click “Install” next to each listed package, such as:

   - ComfyUI-KJNodes

   • Any other required node sets

Once the installation is complete, click “Restart” in the bottom-left corner. When prompted, confirm to reboot the interface. After restarting, the previously red nodes should now appear as active components.

Below, you’ll find all the configuration details needed to get started with generating edited images from your uploaded input using the HiDream-E1-1 model.

We'll start by showcasing the base settings. After that, we’ll walk through a series of examples where both the input image and the image editing prompt vary, demonstrating the model’s flexibility across different scenarios.

Load Diffusion Model Node

• UNet Name: hidream_e1_1_bf16_safetensors

• Weight DType: fp8_e4msfn

Quadruple CLIP Loader Node

Load the following 4 CLIP models:

1. clip_g_hidream.safetensors

2. clip_l_hidream.safetensors

3. llama_3.1_8b_instruct_fp8_scaled.safetensors

4. t5xxl_fp8_e4m3fn_scaled.safetensors

VAE

• VAE File: ae.safetensor

Resize Image V2 Node:

Before editing, upload your input image. Be mindful of resolution to avoid memory errors.

• upscale_method lanczos

• keep_proportion: resize

• pad_color: 0,0,0

• crop_position: center

• divisible_by: 2

• device: cpu

Prompt Configuration

• Edit Prompt: She is smiling. Add dramatic backlighting from a large explosion behind her — fiery orange and red hues casting bold rim light across her body. Add sparks, debris in midair, and swirling snow clashing with heat distortions in the background. Change her handguns to gold color.

Generation Settings

• control_after_generate: random

• cfg_conds: 2.3 (range from 2.0 - 5.0 for best results)

• cfg_conds2_negative: 2.0

• sampler_name: euler

• scheduler: beta

• steps: 28

• denoise: 1.0

In this section, we’ll explore a variety of examples where both the input image and the editing prompts differ, showcasing the model’s versatility across diverse use cases.

Example 1

Edit Prompt: Transform into a vibrant anime cartoon style with bold, clean linework and bright, saturated colors.

Example 2

Edit Prompt: Swap background for higher Gotham vantage point with glowing skyline. Adjust pose to lean subtly forward, emphasizing chest curve and eye contact. Add reflective suit highlights, soft rain and magenta lighting from below.

Example 3

Edit Prompt: Add a rectangular sign held across her hips that reads ‘Street Fighter’ in stylized red lettering. Make sure her hands grip the sign naturally. Enhance the fabric stretch over her chest for subtle sensuality. Add warm spotlight reflections on the sign and her outfit, plus soft confetti or light sparkles in the background.

Example 4

Edit Prompt: She is smiling. Add dramatic backlighting from a large explosion behind her — fiery orange and red hues casting bold rim light across her body. Add sparks, debris in midair, and swirling snow clashing with heat distortions in the background. Change her handguns to gold color.

Example 5

Edit Prompt: Render Wonder Woman laying on creek stones at sunrise in clean, simple black-and-white line art. Use clear, defined outlines with minimal shading. Show soft sunlight streaming across her armor and body using gentle line hatching. Emphasize flowing hair and fabric folds with smooth contour lines. Suggest water reflections and creek texture with sparse line patterns and negative space.

In conclusion, the HiDream image editing model (HiDream-E1-1) is a powerful tool for anyone looking to enhance their image editing capabilities using ComfyUI on Runpod. By following the steps outlined in this tutorial, you can successfully set up your environment, download the necessary files, and configure your system for optimal performance. Remember to keep in mind the model's limitations regarding image size, but also embrace the vast potential it offers for commercial use. Whether you’re creating unique avatars, designing logos, or editing icons, the HiDream model can significantly enhance your workflow. We encourage you to explore its features and experiment with different prompts to see what amazing results you can achieve. Happy editing!