ImageScaleToTotalPixels

ComfyUI has a core node called ImageScaleToTotalPixels, and this is the "advanced" version of it: scale an image so it lands at roughly a target total pixel count - 1 megapixel, 2.25, whatever - rather than at a target side length. What makes it "Adv" is the second knob that core lacks: divisible_by, which rounds the final dimensions to a multiple you choose. That's the setting that saves you from the silent disaster of feeding a sampler a width it wasn't designed for.

The point of total-pixel scaling is simple. Models are trained on images of a certain total resolution, and upscaling by side length (512 → 1024) changes the total by 4x, which is how you accidentally OOM. Scaling to a megapixel target keeps the workload predictable: an 0.8 MP image goes to ~1.0 MP regardless of its aspect ratio. Set 1.0 for SDXL-class work, lower for video, and the node handles the math.

How it works

The math is one line: it computes a scale factor so that W*H*scale² equals megapixels * 1024², then rounds the new width and height up to the nearest divisible_by. Rounding means the actual result is usually slightly over the target - 1.0 MP with divisible_by = 8 can land at 1.04 MP. That's the trade for getting a model-friendly size, and it's the right trade. The resize itself goes through ComfyUI's common_upscale with your chosen method, plus an optional center crop if you set crop_method to center.

Inputs and outputs

• image (IMAGE) - input, single or batch.
• upscale_method - nearest-exact, bilinear, area, bicubic, lanczos. For downscaling, area is the one that handles it properly (anti-aliased); lanczos for quality upscales.
• megapixels (FLOAT, 0.01–64, default 1) - the target total.
• divisible_by (INT, 1–512, default 1) - round dimensions to a multiple. Set 8 or 16 for video/ViT-friendly sizes; 1 disables.
• crop_method - disabled or center.

One output: IMAGE, the resized result.

Installing it

Part of ComfyUI-lhyNodes:

• ComfyUI Manager → search lhyNodes → Install, restart.
• Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/lihaoyun6/ComfyUI-lhyNodes.git
python -m pip install -r ComfyUI-lhyNodes/requirements.txt

Restart ComfyUI.

Where people get burned

The two most common mistakes are both defaults. upscale_method has no default in the schema, so it forces you to pick - and the first thing beginners pick is lanczos, which is heavy and ring-prone for a downscale; reach for area when shrinking. And divisible_by = 1 means no rounding, which quietly reintroduces the odd-sized-output problem this node is supposed to solve - if you're feeding a video model or a ViT-based model, set it to something meaningful like 16. The megapixel target is total, not per-side, so don't be surprised that a 2.25 MP setting doesn't produce a 1500×1500 square; it produces whatever keeps the aspect ratio at that total. For most workflows, set the MP and the divisor and forget it.