Image Dimensions From MP alt (CRT)

This is the sibling of the pack's Image Dimensions From Megapixels, tuned for a different job. Where the original snaps to a 16-grid by default and takes any megapixel float you type, the "Alt" version defaults to a 64-grid and picks your megapixels from a dropdown of preset values - and it additionally hands you a ready-made "W × H" resolution string. If you're feeding LTX or Wan latents, 64 is the multiple those models actually want, so this node is often the one you reach for in video workflows.

Two near-identical nodes for two slightly different habits: the Alt's dropdown forces you into clean presets (0.1 → 2.5 MP in 0.1 steps), which is genuinely nice for repeatability - same target every time, no typos.

How it works

Same math as the original: it takes an image to read the aspect ratio, computes dimensions that hit your target megapixels at that ratio, then snaps both to the multiple_of grid. The differences:

• megapixels is a fixed dropdown of "0.1" through "2.5" in 0.1 steps instead of a free float. No arbitrary 1.37 MP runs - pick from the ladder.
• multiple_of defaults to 64 (range 8–512, stepping by 8), which is the latent-friendly grid for LTX and Wan-style video models.
• It emits a third output, resolution, a STRING like "1216 × 704" - handy when a node or your own logging wants a human-readable size, or you want to display the effective resolution on the graph.

Inputs and outputs

• image (IMAGE) - reference for aspect ratio.
• megapixels - dropdown preset, default "1.0".
• multiple_of - grid to snap to, default 64.

Outputs: width (INT), height (INT), resolution (STRING).

Installing it

CRT-Nodes: ComfyUI Manager → CRT-Nodes, or clone + pip install -r requirements.txt, restart. Under CRT/Image, right next to the original.

Where people get burned

Don't overthink the "Alt." It's not newer or better - it's the preset-and-64 flavor. If you need a 16-grid and precise floats (say, 1.6 MP for an SDXL-sized output), the original is the right pick; if you want video-model-friendly multiples and a resolution string, use this one. The trap is choosing by name rather than by what's downstream. And as with the original, it sizes by the input image's aspect - feed it the right reference or your resolution will inherit the wrong shape.