KfConditioningAddx10

The ten-input version of Add Conditions, and the node the pack's own showpiece workflow is built around. In the README's "prompt entanglement" example - six prompts interleaved simultaneously, each contributing at every frame with a shifting sinusoidal weight - you need to add six weighted conditionings into one. Chaining five two-input Add Conditions nodes would bury the graph. This node takes up to ten CONDITIONING inputs and sums them in a single step.

The README's one requirement mirrors the curve variants: at least cond_0 must be non-empty. The other nine are optional, and that's what makes the node usable - you don't need ten prompts to benefit; you can use it for two, three, or four and leave the rest unplugged.

How it works

All connected inputs are deep-copied and summed - cross-attention tensors added element-wise, pooled outputs added likewise - into a single CONDITIONING output. Unconnected slots contribute nothing, which is the same "optional positions are all fine to leave empty" story the README spells out for the x10 add nodes. Each input is expected to be a single conditioning entry; the node assumes that structure, so feed it plain per-prompt conditionings, not pre-batched ones.

The inputs that matter

• cond_0 (required, forceInput) - the first conditioning. Must be connected.
• cond_1 through cond_9 (optional) - each defaulting to 0 (not contributing) when left unplugged.

The output

CONDITIONING - the summed conditioning for your sampler.

How to install it

Standard pack install: ComfyUI Manager → search ComfyUI-Keyframed, or

cd ComfyUI/custom_nodes
git clone https://github.com/dmarx/ComfyUI-Keyframed

then restart ComfyUI. No model downloads - just the keyframed library (plus toolz), auto-installed on first import.

Troubleshooting

The same two lessons as its two-input sibling, scaled up. Adding doesn't normalize: ten prompts each weighted 1 sum to 10, and the sampler will read a ten-times-strength soup. The entangled-curves workflow avoids this by construction - its sinusoidal weights are engineered so the sum of all curves equals 1 at every moment - so if you're doing your own thing, keep the weights balanced or expect mush. And when the output looks like none of your prompts made it through, check that cond_0 is actually connected; with nine optional slots silently sitting at zero, a missing foundation produces a sum of nothing and the error messages are entirely absent.