This is now the #1 AI image editor. Free & open source

Key Concepts:

• AI Image Editing
• Open-Source Software
• Text-to-Image Editing
• Hydream E1.1
• ComfyUI
• Image Quantization
• Micro-editing
• Style Transfer
• Deep Fakes
• Image Restoration
• Image Colorization
• VRAM (Video RAM)
• CFG (Classifier-Free Guidance)
• Sampling Steps

Hydream E1.1: A New Open-Source AI Image Editor

The video introduces Hydream E1.1, a new free and open-source AI image editor, as a superior alternative to previous tools like Omni Genen 2 and Flux Context Dev. It highlights Hydream E1.1's position as the top-ranked open-source image editor on the Artificial Analysis leaderboard, surpassing Flux Context Dev by over 60 ELO points. The video aims to demonstrate its capabilities and provide a step-by-step installation guide.

Capabilities and Examples:

Hydream E1.1 allows users to edit images using text prompts. Examples include:

• Framing a person as a masterpiece in an art museum.
• Changing the background of an image to a dark server room and making text glow.
• Removing bullets from a scene.
• Transforming an image into an underwater scene with a sea turtle.
• Changing the color scheme of a design to black while preserving details.
• Modifying a person's clothing, appearance, and background while maintaining their pose and facial features.

Personal Tests and Results:

The video showcases personal tests to evaluate Hydream E1.1's performance:

• 3D Pixar Style: Transforming an image into a 3D Pixar style, taking 2-3 minutes on a GPU with 16GB VRAM using 20 steps.
• Color Change and Background Addition: Changing a car's color to blue and adding mountains in the background, preserving details like grass blades and the driver.
• Object Insertion: Adding Mount Fuji to a painting while keeping the rest of the photo intact.
• Text Editing: Changing the text "joker" to "clown" on a movie poster, with some loss of detail and font consistency.
• Focus Adjustment: Focusing on a blurry flower in the foreground and blurring the background.
• Tattoo Removal: Removing tattoos from a person's arms, with a slight increase in image saturation.
• Raindrop Removal: Removing raindrops from an image while maintaining details.
• Micro-editing a Person: Changing a woman's outfit to a cowboy hat and red dress and changing the background to a snowy forest, preserving her face.
• Object Removal: Removing objects from a table while leaving objects on a window, demonstrating prompt understanding.
• Anime Transformation: Turning a photo into an anime style, preserving details and pose.
• Photo Restoration and Colorization: Restoring and colorizing an old, damaged photo, adding the prompt "as if it was a modern-day photo" for better colorization.

Installation Guide Using ComfyUI:

The video provides a step-by-step guide to install Hydream E1.1 using ComfyUI:

1. Prerequisites: ComfyUI must be pre-installed.
2. Download Required Files:
   • Text Encoders: Download all files from the text encoders folder (clip L safe tensor file, llama 3.1). Place these files in the ComfyUI/models/text_encoders folder.
   • VAE: Download the AE.safe tensors file from the VAE folder. Place this file in the ComfyUI/models/VAE folder.
   • Hydream E1.1 Model: Download the Hydream E1.1 model from the diffusion models folder. Due to the large size (32GB), quantized versions (Q2, Q3, Q4, etc.) are recommended for users with less VRAM. These are provided by ND 911. Place the downloaded model in the ComfyUI/models/diffusion_models folder.
3. Start ComfyUI: Double-click to start ComfyUI.
4. Update ComfyUI: Click on "manager" and press "update Comfy UI" to ensure the latest nodes and dependencies are installed. Restart ComfyUI after updating.
5. Load Workflow: Download the workflow file provided by ND 911 and drag and drop it onto the ComfyUI interface.
6. Select Models: In the ComfyUI workflow, select the downloaded models for each corresponding node (clip G hydream model, clip L hydream, T5XXL scaled, llama 3.18B model, AE.safetensors).
7. Enter Prompt and Upload Image: Enter the desired text prompt and upload the image to be edited.
8. Adjust Settings (Optional):
   • CFG: Controls how literally the AI follows the text prompt.
   • Image CFG: Controls how literally the AI follows the image details.
   • Sampler Name and Schedule: Algorithm used to generate the image (default: Euler Simple).
   • Steps: Number of iterations (default: 20, a balance between speed and quality).

Technical Terms and Concepts:

• VRAM (Video RAM): Memory on the graphics card used for processing images and videos.
• Quantization: Reducing the size of a model by compressing its parameters, allowing it to run on GPUs with less VRAM.
• CFG (Classifier-Free Guidance): A parameter that controls how closely the AI adheres to the text prompt.
• Sampling Steps: The number of iterations the AI performs to generate an image. More steps generally result in higher quality but take longer.

Comparison with Flux Context Dev:

Hydream E1.1 is considered better for micro-editing, while Flux Context Dev is slightly better for style transfer and editing the entire image. Both are leading free and open-source image generators.

Conclusion:

Hydream E1.1 is a powerful open-source AI image editor that excels at micro-editing and offers impressive capabilities for image manipulation and restoration. While it may be slower than Flux Context Dev for certain tasks, it remains a valuable tool for users seeking free and open-source alternatives. The video encourages viewers to experiment with Hydream E1.1 and report any installation issues in the comments.