512

Five hundred twelve. The resolution that built a generation of image models. SD 1.5 was trained at 512×512, which is why 512 is everywhere in that lineage's workflows - from base resolutions to the latent size to the "small" end of an upscale ladder. This constant node from 2daadv's ComfyUI-GadgetNodes pack gives you 512 as a hardwired INT output that cannot be accidentally changed, and honestly, that's the whole story. It's one of seven constants in the Gadget/core/const family, all built the same way: no inputs, one fixed output, zero state.

Mechanically there's nothing to trip over - empty required inputs, a single INT output of 512, no frontend extensions, no dependencies beyond the pack's base install. The pack itself (MIT, one developer, brand new as of mid-2026 with no community reputation either way) installs in one step:

cd ComfyUI/custom_nodes
git clone https://github.com/2daadv/ComfyUI-GadgetNodes.git
pip install -r ComfyUI-GadgetNodes/requirements.txt

Restart, and it's under Gadget/core/const - or just search "GadgetNodes" in ComfyUI Manager.

Where you actually wire a 512:

• SD 1.5 workflows. If you're running a 1.5-lineage model, 512 is your native resolution. Locking it in with a constant keeps the base size honest while every other knob in the graph gets played with. This matters more than it sounds - generating 1.5 models at 768 is where anatomy starts lying to you.
• The middle rung of a halving ladder. 1024 → 512 → 256 is the canonical progressive-upscale shape, and 512 is the rung where the detail work actually happens.
• Any widget default you want to reset to. Some samplers and tile nodes ship with sane defaults that you keep fighting; a constant on top is a way to say "no, 512, always."

The honest caveat is the same one I gave for 256: 512 is only "the right number" if your model and your intent say so. On SDXL and Flux the native square is 1024, so a locked 512 there is a choice, usually a fast-pass or VRAM-saving one - fine, but make it on purpose. And remember that on modern architectures the model decides the ideal footprint, not nostalgia (concepts.md is blunt about this: what fit in 512 on SD 1.5 is not what fits on an LLM-encoder model).

One source, several consumers: that's the entire reason this node exists. Wire the 512 output into every input that wants it, and stop retyping it.