chx_input_data

Apt_Preset really wants you to use its own all-in-one loader, sum_load_adv, which per the README handles checkpoints, UNet+clip combos for Wan/Flux/SD3.5/HiDream, and preset saving all in one node. But you don't always want that - maybe you already have a CheckpointLoader you like, or you're mid-migration and only want to try Apt_Preset's samplers without rebuilding your loading stage. chx_input_data is the adapter for exactly that situation: it takes the four things every ComfyUI graph already produces - model, latent, VAE, clip - and packages them into the pack's context wire so the rest of Apt_Preset's controller and sampler nodes will accept them.

Why the pack needs an adapter node at all

Apt_Preset's whole design, per its own README, is built around a "Loader → Controller → Sampler" chain connected by a bundled RUN_CONTEXT type instead of four or five separate wires - the same idea rgthree's Context nodes apply to general ComfyUI graphs, just scoped specifically to this pack's sampler family. That's genuinely nice once you're all-in on Apt_Preset, but it means the pack's samplers won't take a raw MODEL or LATENT input directly - they expect context. If your workflow starts with vanilla ComfyUI loaders instead of sum_load_adv, chx_input_data is the node that gets you from "normal ComfyUI outputs" to "something Apt_Preset's samplers will accept."

Inputs and outputs

There's no configuration here - it's pure plumbing:

• model, latent, vae, clip - all required, all standard ComfyUI types. Wire these straight from your existing checkpoint/UNet loader, your CLIP loader, your VAE loader, and whatever produced your starting latent (an EmptyLatentImage, a VAE Encode, etc.).

Outputs mirror the inputs, plus the new bundle:

• context - the packaged RUN_CONTEXT, ready to feed into basic_Ksampler_simple, chx_IPA_basic, or any other Apt_Preset controller/sampler node.
• model, latent, vae, clip - passed straight through, so you can still tap any of them individually further down the graph without unpacking the context again.

Installing it

Through ComfyUI Manager: search ComfyUI-Apt_Preset, install, restart. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

then run install.bat (Windows) for dependencies, restart ComfyUI. This node has no model files or extra dependencies beyond the base pack install - it's a type conversion, nothing more.

Common issues

A downstream Apt_Preset node still rejects the connection. Double check you're wiring the context output of this node, not one of the passthrough model/latent/vae/clip outputs, into the next node's context input. It's an easy slip since all five outputs sit right next to each other on the node.

You built your latent with the wrong dimensions for the model. This node doesn't validate that your latent, model, and VAE are actually compatible with each other (SD1.5 latent into an SDXL model, for instance) - it just bundles what you give it. If your sampler run produces garbage, check that mismatch first; it's a stock ComfyUI failure mode that carries straight through here.

You're missing one of the four inputs. All four - model, latent, vae, clip - are required with no optional fallback. If your loader setup doesn't cleanly produce all four as separate outputs (some all-in-one loaders bundle VAE into the checkpoint output differently), you may need an extra VAELoader or similar stock node to fill the gap before this one will accept the connections.