Into Flake Data (Select)

IntoFlakeDataSelect is the surgical version of the Flakes on-ramp. Where IntoFlakeDataAll builds a flake_data bundle from scratch, this one starts with an existing bundle, lets you override whichever individual fields you connect, and passes everything else through untouched. "Only connected pins override the original data" is the whole contract, and it's a good one - it means you can reach into a preset-plus-stack pipeline and change exactly one thing without rebuilding the graph.

The dropdown and [+] button let you add only the input pins you need, same idea as FlakeDataSplitSelect but for inputs. The menu covers the full bundle: model, clip, vae, positive, negative, latent, filename_prefix, width, height, steps, cfg, sampler_name, scheduler. And because the underlying override code is field-wise (not chunk-wise like the "All" builder), connecting just vae or just cfg works exactly as you'd hope - that one field swaps, the rest of the bundle survives.

This is the node you reach for when a preset is 95% right. The checkpoint and prompts are perfect; you just want this render at 1216×1216 instead of the preset's 832×1216. Or you want to A/B a scheduler against the preset's default without editing (or cloning) the YAML. Or you want to swap in a different VAE for one pass. Each of those is a single pin, connected and disconnected as needed. There's a useful subtlety on filename_prefix: setting it overrides the "preset" part of the output filename - the node stores it in the filename state, so downstream Save Image / Flake Generate will use your value as the prefix base rather than the preset's name.

It's also a tidy middle step when you're mixing ecosystems. Feed it a bundle from a Flake Model Preset, override the model with something from your own loader, and you've got a hybrid workflow that uses the pack's prompt/conditioning machinery but your own checkpoint - no preset file required.

Inputs and outputs. Required: flake_data. Optional: any subset of the thirteen fields, added via dropdown and [+]. Output: flake_data with the connected overrides applied.

Install. ComfyUI Manager → search ComfyUI Flakes → Install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/JeyzerMC/comfyui-flakes

then restart. PyYAML is the only dependency and it ships with ComfyUI core.