Ard 4Img Combine

Ard 4Img Combine is the reassembly half of this pack's four-tile workflow. You crop a latent into quadrants, generate or upscale each one separately, decode them, and then this node pastes the four images back into a single canvas. If you've ever done a tiled upscale by hand in an editor, this automates the "put them back together" step.

The inputs: image_1 and image_2 are required, image_3 and image_4 optional, and there are two dropdowns - oreintation (yes, misspelled, that's the actual parameter name) with over / beside, and input_images with 1 and 2 only / 4 images. With "4 images" selected, the layout is a 2×2 grid: image_1 top-left, image_2 top-right, image_3 bottom-left, image_4 bottom-right. With "1 and 2 only", beside lays them side by side and over stacks them vertically. Canvas size is computed from the images rather than set by you, which keeps things simple - your tiles should already be the same dimensions for the 2×2 case.

Outputs are two: Combined Image (the paste result) and Image Mask. Here's the honest warning: that mask output is mostly a dummy. The code builds it the same way ComfyUI's Load Image builds alpha masks, and a plain RGB composite has no alpha channel - so you get a blank 64×64 zero tensor pretending to be a mask. Don't build anything that depends on it. If you need a real selection mask, generate one yourself.

A few gotchas worth knowing before you wire it up:

• The node is flagged as an output node, so the combined result shows up in the image panel - good, that's what you want for the final stitch.
• oreintation and input_images interact: for the 2×2 case the code pastes the right column using image_2's width/height in one spot and image_3's in another. If your four tiles aren't uniform, the bottom-right paste can land slightly off. Keep tiles the same size.
• It operates on pixel IMAGE tensors, not latents - decode first. The pack's Ard 4Vae Decode is the intended partner for turning four latents into four images to feed in here.

When would you actually use this? When you've split a big image into quadrants, detailed each tile at higher resolution, and want them back on one canvas - the poor-man's tiled upscale. If you're only stitching two images occasionally, the built-in Image Concatenate nodes do the same job with fewer surprises; this node's value is the 2×2 convenience and the fact that it lives in the same pack as the rest of the tile pipeline.

Install is the standard pack route - ComfyUI Manager → search ComfyUI-Ardenius, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ArdeniusAI/ComfyUI-Ardenius

then restart ComfyUI. The pack's requirements.txt pulls in civitai and moviepy; a stray import error at startup is the Save Image node asking for comfyui_controlnet_aux, not this one.