Load Image Resize (CRT)

Load Image Resize (CRT) is the standard ComfyUI LoadImage with a resizing step welded on. You pick an image from the upload dropdown, give it a target size in megapixels, and it loads and resizes in one go - with the dimensions snapped to a multiple of your choice so they're always VAE- and attention-friendly. It's a small convenience that removes a ImageScale node (and its "wait, what size is this now?" confusion) from half of your graphs.

The size inputs are the whole point. megapixels (default 1.0) is the target area of the output - 1.0 MP means roughly 1024×1024, 0.5 MP about 707×707. The node keeps the source aspect ratio, so the actual width and height derive from your target area. multiple_of (default 16) rounds each dimension up to a clean multiple - and this is the bit beginners skip and then regret. Diffusion latents live at 8× downsampling, and a lot of models are happiest when both dimensions divide by 16 (or 64 for some architectures). If you feed a sampler a 1001×1001 latent you'll get artifacts or outright errors depending on the model; snapping to 16 keeps you out of that ditch for free.

It outputs four things: IMAGE (the resized pixels), MASK (the alpha/transparency channel, present even for images without alpha - it's just all-white, same as core LoadImage), plus width and height as INTs. Those last two are quietly the most valuable. Wire them into a resolution calculator, a text label, or an empty-latent node and you never have to guess what size the pipeline actually resolved to. The image dropdown reads from ComfyUI's input folder and lets you upload a new file right from the node, exactly like the built-in loader.

Where this shines: img2img at a fixed working resolution, hi-res-fix loops where the first pass must land on a clean size, batch pipelines that must feed every image in at identical dimensions, and any workflow where the source files vary wildly in size but your model doesn't. Because the resizing happens at load, everything downstream sees a consistent shape - no mid-graph ImageScale drifting.

The setup is the pack's: ComfyUI Manager → search "CRT-Nodes", or clone https://github.com/plugcrypt/CRT-Nodes.git into custom_nodes and restart. It uses only core ComfyUI image loading and a resize, so there are no extra model downloads. One habit worth building: if you're loading a reference for an edit model, don't over-shrink it - a target MP that leaves you below ~256px a side loses the details you were trying to keep. Otherwise, default of 1.0 MP and multiple-of-16 covers most cases cleanly.