Into Flake Data (All)

IntoFlakeDataAll is the on-ramp to the Flakes ecosystem: it takes ordinary ComfyUI values - the kind every other node in the world produces - and bundles them into a flake_data wire. The Flake Stack, Flake Combo, and Flake Generate nodes all speak flake_data, and this is how you talk to them without using the pack's own preset loaders. You can keep your favorite checkpoint loader, your hand-tuned conditioning, your own empty-latent setup, and just wrap the results.

Every input is optional, and that's the design: "only connect the ones you need" is literally the node's description. The model trio works together - connect model, clip, and vae and they're packed as a unit. Connect positive and negative conditioning and the prompt half is done. Then the sampler values are plain widgets with sensible defaults: width/height 1024×1024, steps 20, cfg 7.0, sampler_name euler, scheduler normal. Leave them alone and that's what a downstream Flake Generate will use.

The one structural gotcha is that the bundle is assembled in chunks, not field-by-field. If you connect model and clip but not vae, the pack stores no model bundle at all rather than a half-baked one - so a downstream Flake Generate that expects a model will quietly have nothing to sample with. Same for conditioning: positive without negative means neither is packed. So the rule is: connect all three of the model trio, or none; connect both conditionings, or neither. If you only want to change one value of an existing bundle, that's what IntoFlakeDataSelect is for - it starts from an incoming flake_data and overrides only the pins you connect.

The realistic use: you've been asked to slot a Flake Stack into a workflow that was built the old-fashioned way. Wire your existing checkpoint loader's outputs into model/clip/vae, your existing CLIP Text Encode results into the conditionings, and suddenly the whole preset-stacking machinery - the grid UI, the LoRA stacking, the variants - is available to your graph. It's also the honest way to evaluate the pack: try it on your current setup before committing to building everything around model presets.

Inputs and outputs. Thirteen optional inputs (model, clip, vae, positive, negative, latent, width, height, steps, cfg, sampler_name, scheduler), one flake_data output.

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.