AI Enhancement Batch

This node combines two of the pack's simpler enhancement ideas - Real-ESRGAN upscaling and perceptual-color-space correction - into one batch-friendly node, with the option to chain them. If you're processing a folder of images and want either or both applied consistently without stitching together separate nodes per image, this is the one built for that.

How it works

enhancement_method picks the actual job: real_esrgan runs the same upscaling model as this pack's standalone RealESRGANNode, perceptual_color adjusts color and contrast working in a perceptually uniform color space instead of raw RGB, and combined runs both in sequence. The color-space choice matters more than it sounds like: RGB and even HSV don't match how humans actually perceive brightness and hue changes, so adjustments made directly in RGB tend to introduce unwanted hue shifts as saturation changes. Working in a space like Oklab (the default here) keeps perceived hue and lightness more stable while you push contrast or saturation, which is exactly why newer color tools have been moving toward these spaces instead of plain RGB math.

The inputs and outputs that matter

• images - required, plural - this is a batch node.
• enhancement_method - real_esrgan, perceptual_color, or combined.
• batch_size (1–16, default 4) - how many images process at once; higher needs more VRAM.
• real_esrgan_model (optional, default RealESRGAN_x4plus) - same three variants as the standalone node: general photos, anime/illustration, or a lightweight general option.
• color_space (optional, default oklab) - oklab, jzazbz, or the older lab. Oklab is the newer, generally recommended default; the others are there if you have a specific reason to want them.
• enhancement_strength (optional, 0–2, default 0.7) - only relevant when perceptual_color or combined is doing color work.
• Outputs: enhanced_images and processing_info.

How to install it

Search Eric_Image_Processing_Nodes in ComfyUI Manager, or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/EricRollei/Eric_Image_Processing_Nodes
cd Eric_Image_Processing_Nodes
pip install -r requirements.txt

Restart ComfyUI. The Real-ESRGAN weights auto-download on first use exactly as they do for the standalone node - if you've already run RealESRGANNode once, this one reuses the same cached weights rather than downloading again. The perceptual-color path is pure math and needs nothing extra beyond the pack's core requirements.

Common issues & troubleshooting

Running out of memory on a batch. Drop batch_size first - it's the biggest memory lever here, especially in combined mode where both the upscaler and the color correction are running per image.

Color shifts look unnatural after perceptual_color or combined. Try enhancement_strength lower before switching color_space - 0.7 is already a moderate default, and pushing toward 2.0 will exaggerate any hue drift that's present. If you have a specific reason to distrust Oklab for your content, jzazbz and lab are the fallbacks.

Anime source looks worse after real_esrgan. Same mismatch as the standalone node - set real_esrgan_model to the anime variant, not the general-photo default, when your content is illustration rather than photography.

Combined mode is slow. You're running two separate techniques per image in sequence, on top of batching - this is inherently the heaviest option this node offers. If you only need one of the two effects, use real_esrgan or perceptual_color alone rather than paying for combined by default.