🔗 Attention Modifier Concat

Every node in the pack's attention-modifier family outputs the same thing: a custom ATTNMOD wire that is really a list of per-block instruction dicts. Attention Modifier Concat exists for the moment when one modifier isn't enough - it takes two ATTNMOD wires in (parameters_1, parameters_2) and appends the second list to the first, handing back one combined wire.

That's the entire node. Two inputs, one output, no settings, no sigma windows, nothing to tune. If you've seen the pack's Attention Modifier Parameters or Bruteforce nodes, you already know the drill: those build the list, Concat merges two lists, and the Advanced CFG Controller (Expert) node applies the result. When you want to patch a self-attention quirk and a cross-attention behavior in the same run, you don't need two Expert inputs - you build both modifiers, join them here, and feed the single wire into one attention_modifiers_positive slot.

Ordering is the only thing to think about. Concat is parameters_1 + parameters_2, left to right, and because each entry carries its own sigma window and block coordinates, order rarely matters - they don't collide unless you point two entries at the same block. If you do, the later one wins, so put the more important patch second.

The honest framing: this is plumbing, and it's thin. If you're building one modifier at a time, you'll touch Concat only when a workflow asks you to stack patches - it's the "maybe one day" node in the drawer. The one genuinely useful habit: because ATTNMOD is an opaque custom type, keep a text node on the Parameters as string outputs of your modifiers and let Concat's combined output flow to the Expert. It won't tell you more than the inputs did, but it's how you sanity-check a stacked chain in a pack that's young enough to still have rough edges. As with everything here: no models, no extra dependencies - install the pack once, find it under model_patches/Advanced_CFG_Controller/attention_modifiers, and remember the Expert node has to be in the graph for any ATTNMOD to actually change anything.