This new AI image editor is a BEAST

Key Concepts

• Quen ImageEdit: A free and open-source AI image editor.
• Flux Context Dev: A leading open-source image editor used for comparison.
• Microediting: Precise and localized image manipulation using text prompts.
• Background Swapping: Replacing the background of an image with a different scene.
• Model Sheet Generation: Creating front, back, and side views of a character from a single image.
• ComfyUI: A popular platform for generating AI images and videos offline.
• Diffusion Model: A type of machine learning model used for image generation.
• LoRA (Low-Rank Adaptation): A technique to fine-tune pre-trained models with less computational resources.
• Text Encoder: A component that converts text prompts into a format understandable by the AI model.
• VAE (Variational Autoencoder): A type of neural network used for encoding and decoding images.
• K Sampler: A component in ComfyUI that generates images based on a seed, steps, CFG, and sampler algorithm.
• Seed: The starting point for image generation, influencing the specific output.
• Steps: The number of iterations the AI model performs during image generation.
• CFG (Classifier-Free Guidance): A parameter that controls how closely the AI follows the text prompt.
• GGUF: A file format for quantized models, allowing them to run on systems with lower VRAM.
• Quantization: Reducing the precision of model weights to decrease memory usage and computational requirements.

Quen ImageEdit: A Powerful and Free AI Image Editor

Quen ImageEdit is introduced as a new, free, and open-source AI image editor that rivals existing solutions like Flux Context Dev. The video highlights its capabilities through various examples and comparisons, demonstrating its strengths in image manipulation, microediting, and understanding complex prompts.

Demos and Capabilities

• Consistent Character Generation: Quen ImageEdit can generate multiple images of a character from a single reference image, maintaining consistent facial features, clothing, and details across different poses and actions. Example: Generating illustrations of a Capy Barra character doing various activities.
• View Generation: The tool can generate different views (front, side, back) of an object or character from a single image. Examples: Generating front views of a cow and a person, side views of a person, and back views of a crow and a lion.
• Microediting with Text Prompts: Users can make precise edits to images using text prompts without manual brushing or editing. Examples: Replacing a design on a t-shirt, adding a sign to a photo, removing a strand of hair, and changing the color of a single letter.
• Background Swapping: The tool can seamlessly replace the background of an image while maintaining realistic lighting and shadows. Example: Changing the background of a photo to a beach or a classroom.
• Clothing Manipulation: Quen ImageEdit can change the clothes of people in photos.
• Text Editing: The tool can edit text in images while preserving the original style and font. Examples: Changing "health insurance" to "financial planning" or translating text to Chinese.
• Object Replacement: Characters in a poster can be replaced while preserving the original design.

Comparison with Flux Context Dev

The video compares Quen ImageEdit with Flux Context Dev, a leading open-source image editor, through several tests:

• Microediting and Background Swapping: Quen ImageEdit more accurately generated a post-apocalyptic background and successfully put a subject in a red bikini, while Flux Context Dev seemed more censored and less accurate in background generation.
• Color Correction: Quen ImageEdit successfully corrected the colors of a deliberately distorted photo, while Flux Context Dev failed to produce a satisfactory result.
• Deblurring: Both editors performed well in deblurring a deliberately blurred image.
• Zooming and Detail Enhancement: Quen ImageEdit produced a much sharper and more detailed zoomed-in image of a bird compared to Flux Context Dev.
• Photo Restoration: Both editors performed well in restoring a damaged old photo, with Flux Context Dev being more accurate in portraying the face, but Quen ImageEdit making the photo look more modern.
• Model Sheet Generation: Quen ImageEdit successfully generated front, back, and side views of a character, while Flux Context Dev failed to generate the back view.
• Text Translation: Quen ImageEdit accurately translated text to Chinese while preserving the font, while Flux Context Dev produced gibberish.
• Prompt Understanding: Quen ImageEdit accurately placed a woman in front of the text "Quinn" but behind "ImageEdit," while Flux Context Dev failed to do so.
• Pattern Addition: Both editors successfully added cheetah, polka dot, and iridescent patterns to text, with Quen ImageEdit's patterns being subjectively preferred.
• Watermark Removal: Quen ImageEdit successfully removed watermarks without altering the image's contrast, while Flux Context Dev slightly darkened the image.
• UI Removal: Both editors removed the user interface from a gameplay scene, with Flux Context Dev producing a more detailed and color-accurate result.
• Style Transfer (Anime): Both editors converted a photo to anime, but with different styles.
• Style Transfer (Lego): Quen ImageEdit produced a more convincing Lego-style transformation compared to Flux Context Dev.
• Depth Map Generation: Quen ImageEdit generated a reasonably accurate depth map, while Flux Context Dev failed.
• Style Transfer (Watercolor): Both editors produced good watercolor paintings, with Quen ImageEdit's result being subjectively preferred.

Online and Offline Usage

• Online: Quen ImageEdit can be used online through a chat interface (chat.quen.ai), similar to ChatGPT, with an image editing feature. This method may have daily limits and censorship.
• Offline (ComfyUI): The recommended method is to run Quen ImageEdit offline using ComfyUI for unlimited and uncensored usage.

Installation and Setup in ComfyUI

The video provides a step-by-step guide to installing and setting up Quen ImageEdit in ComfyUI:

1. Download Models: Download the required files (Quinn imageedit diffusion model, LoRA, text encoder file, and VAE) from the provided links and place them in the appropriate ComfyUI folders (models/diffusion_models, models/loras, models/text_encoders, and models/vae, respectively).
2. Update ComfyUI: Update ComfyUI to the latest version using the manager.
3. Download Workflow: Download the pre-built ComfyUI workflow JSON file.
4. Load Workflow: Drag and drop the JSON file onto the ComfyUI interface.
5. Select Models: In the workflow, select the appropriate models (Quinn image edit, Quinn 2.5VL, and Quinn image VAE) in the dropdown menus.
6. Enable Lightning LoRA (Optional): To speed up generation, unbypass the LoRA component (Ctrl+B) and select "Quinn image lightning four steps." Set the step count to 4 and CFG to 1.
7. Upload Image and Enter Prompt: Upload the image to be edited and enter the desired prompt.
8. Run Workflow: Click "Run" to generate the edited image.

Low VRAM Usage

For users with low VRAM, the video explains how to use quantized versions of Quen ImageEdit:

1. Download Quantized Model: Download a quantized version of the model (e.g., Q2) from the provided link and place it in the ComfyUI/models/unet folder.
2. Install GGUF Nodes: Install or update the "comfy_gguf" custom node in ComfyUI using the manager.
3. Replace Load Diffusion Model Node: Replace the "Load Diffusion Model" node with a "UNet Loader GGUF" node.
4. Select Quantized Model: Select the downloaded quantized model in the dropdown menu of the "UNet Loader GGUF" node.
5. Run Workflow: Click "Run" to generate the edited image. Note that quantized models may result in lower image quality.

Data Impulse Sponsorship

The video includes a sponsorship message for Data Impulse, a service providing access to over 90 million IPs across 195 countries for web scraping, tool testing, and automation. It highlights the service's flexibility, control over IP rotation, compatibility with SOCK 5 and HTTPS, and a user-friendly mobile dashboard.

Conclusion

Quen ImageEdit is presented as a highly capable and versatile AI image editor that excels in various tasks, including microediting, background swapping, and style transfer. Its open-source nature and free availability make it an attractive alternative to existing solutions. The video provides a comprehensive guide to installing and using Quen ImageEdit, both online and offline, with specific instructions for users with low VRAM systems. The comparison with Flux Context Dev highlights Quen ImageEdit's strengths in prompt understanding and accurate image manipulation.