Recall Conditioning

Recall Conditioning reads back a CONDITIONING your loop stashed on a previous run - the encoded-prompt counterpart to Recall Int and Recall Float. It's the node you use when your Cyclist loop is building up a prompt in conditioning space and you want each iteration to continue from the version the last one produced, rather than re-encoding from text every time.

When it's the right tool

Cyclist loops by re-queuing the workflow, and each run forgets everything. Recall Conditioning is how an assembled conditioning survives that gap. The pack's "ugly animal crossbreeds" example is the poster child: every pass adds a new conditioning for a random animal word, and Recall Conditioning brings back the accumulated blend so the next pass adds to it instead of starting over. Do that for a few iterations and you get a prompt that's been composed across many runs - a chimera built one animal at a time.

Be clear-eyed about when not to use it, though. If your loop just changes prompt text, memorizing a string and re-encoding it is simpler and more predictable. Recall Conditioning is specifically for when you're doing conditioning-space operations - concat, combine, average - inside the loop and need to carry the tensor itself forward.

The inputs that matter

• loop_id (STRING, default ForLoop_1) - the memory slot; has to match the Memorize Conditioning that wrote it.
• fallback (CONDITIONING, optional) - used on the first iteration when nothing's stored yet. Wire in your base/starting conditioning so pass one has something to work from; without a fallback, an empty recall has nothing to hand you.

Output is a single CONDITIONING - feed it into your sampler, or into another conditioning-combine node to keep building.

Install

• ComfyUI Manager: search comfyui-cyclist, install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/Pos13/comfyui-cyclist, restart.

Pure Python - nothing to download.

Heads-up

This is memory-only - stored in RAM, not on disk - so a ComfyUI restart clears it and Recall Conditioning falls to its fallback. There's no disk-persistent version, because a CONDITIONING is a live tensor rather than a file; if you need to rebuild it after a restart, you re-derive it from a saved string or image.

The bigger caveat is version fragility. Conditioning is one of the internals ComfyUI changes between releases, and this pack is archived and unmaintained - the author has said they're done supporting it. Plain Int/Float recall will likely keep working for ages; conditioning recall is the node most exposed to a future breaking change. It works today, but validate it on your ComfyUI version before you lean on it, and always wire a fallback so a miss degrades gracefully instead of crashing.