Introduction:
Welcome to this comprehensive guide on using the InstantIR Workflow for automatic blind image restoration. This user-friendly workflow allows you to restore and enhance images with a simple click-and-run setup, eliminating the need for manual configurations or prior knowledge about image degradation. Equipped with state-of-the-art models, this workflow can intelligently correct blurring, remove noise, and boost resolution for sharper, more detailed results.
Workflow:
Step 1. Plugin Installation and Model Download
Start by installing and downloading the plugins and models.
- In the ComfyUI Manager Menu, search for the ComfyUI_InstantIR_Wrapper plugin and install it.
- Download the InstantIR model and place it in the following path: Storage> models> InstantIR> models
- Additionally, download the facebook/dinov2-large model and store it in Storage> models> InstantIR> dino
- Finally, download the latent-consistency/lcm-lora-sdxl model and place it in the following path: Comfyui>models>loras
Step 2. Configure the Workflow Setup
The InstantIR_Loader node is responsible for loading various model checkpoints and configurations for the image processing workflow. Here, you will set up all the models downloaded in Step 1, including the InstantIR mode, dino model, and LoRA.
When selecting the SDXL checkpoint model, it is recommended to choose the Juggernaut model, as it has the ability to generate higher resolution and more detailed images. Additionally, Juggernaut is typically optimized, resulting in relatively faster generation speeds, especially when compared to other high-end generation models.
Step 3. Upload Image for Restoration
After setting up the InstantIR Loader, upload a blind image that you wish to restoration. In this workflow, we'll take below image as reference.
After uploading, set the image's maximum width and height to 1024, and minimum to 0.
Step 4. Enter Image Clarity Prompts
Finally, in the InstantIR Sampler section, input both positive and negative prompts that use to describe the clarity of the image or the outcome of the final objects.
Positive prompts: Photorealistic, highly detailed, hyper detailed photo - realistic maximum detail, 32k, ultra HD, extreme meticulous detailing, skin pore detailing, hyper sharpness, perfect without deformations, taken using a Canon EOS R camera, Cinematic, High Contrast, Color Grading.
Negative prompts: blurry, out of focus, unclear, depth of field, over-smooth, sketch, oil painting, cartoon, CG Style, 3D render, unreal engine, dirty, messy, worst quality, low quality, frames, painting, illustration, drawing, art, watermark, signature, jpeg artifacts, deformed, lowres
Step 5. Final Output of Blind Image Restoration
The restored images will show remarkable improvements in clarity, detail, and overall quality. This advanced process removes noise, corrects blurring, and enhances resolution without needing input on the original issues.
Conclusion
By leveraging state-of-the-art models, InstantIR intelligently enhances textures, sharpens edges, and revitalizes colors, resulting in crisp, high-quality images that retain natural details and visual integrity. This workflow is particularly effective for enhancing low-resolution, compressed, or degraded images, making them appear clear and refined.