pre_Kontext_mul_Image

This was Apt_Preset's preset for feeding up to four reference images into a Flux Kontext-style instruction edit at once - turnaround sheets, "combine these elements into one scene," multi-subject composites, the stuff the README's changelog calls out under "General Control Kontext Stack: Multi-image reference." Like its single-image sibling pre_Kontext, it's now filed under the pack's own "Deprecated" category, so both the one-image and four-image Kontext presets in this pack have been superseded, most likely by a newer combined node rather than one replacing the other.

What it did

Kontext consumes reference images in-context rather than through a separate adapter, which is what made it strong at multi-image composition in the first place - feeding several images in isn't a hack, it's the architecture working as intended. This node exposed up to four image slots plus two parameters that controlled how those references got combined internally.

reference_latents_method offered three strategies: offset and index are two different ways of positioning multiple reference latents relative to each other before the model processes them, and uxo/uno is very likely a reference to UNO, a multi-reference technique used in some Flux-family tooling for combining several subject images into one coherent generation - take that identification as a reasonable read of the label rather than a confirmed detail, since the node itself doesn't document it further. guidance (0–100, default 3.5) is the same classifier-free-guidance-style scale Flux and Kontext expose generally - 3.5 is the standard Flux guidance default, not something specific to this node.

Inputs and outputs

• context (required) - the RUN_CONTEXT chain.
• reference_latents_method (required, enum) - offset, index, or uxo/uno.
• guidance (0–100, default 3.5) - the Flux-style guidance scale.
• image1 / image2 / image3 / image4 (all optional IMAGE) - up to four reference images.

Outputs: context and positive (CONDITIONING) - notably no latent output here, unlike the single-image pre_Kontext, so this one was meant to feed conditioning into a separate latent/image pipeline rather than producing one itself.

What to use instead

Don't build new workflows on this one - it's deprecated, alongside pre_Kontext. Check the pack's current active nodes for whatever now handles multi-image instruction editing; the README's changelog references ongoing consolidation into a broader Kontext control stack that this preset most likely fed into before being retired.

Installing the pack

ComfyUI Manager: search "ComfyUI-Apt_Preset", install, restart. Manually:

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

Restart afterward, then run install.bat on Windows or the equivalent pip install from the repo's requirements file on Linux/Mac.

Common issues

Only some of the four images seem to have an effect. Check which images are actually connected - all four are optional, and depending on reference_latents_method, an unconnected slot is simply skipped rather than treated as an error.

Not sure what uxo/uno actually changes. This is the least-documented parameter on the node, and the pack itself doesn't spell it out beyond the option label. If your result looks off with one method, trying the other two is a reasonable diagnostic step given the node offers no other guidance on when to pick which.

Node missing from the normal node search. It's under the pack's deprecated category - browse there directly if you're trying to open an old workflow that references it, but plan to migrate off it for anything new.