Lora Randomizer (LoraManager)

When you want variety rather than a fixed look - surprise style combos, a batch that isn't all the same LoRA at the same strength - this is the node. Lora Randomizer picks LoRAs at random from a set you give it and emits them as a LORA_STACK, so each run can pull a different combination. It's the exploration tool of the pack: great for finding pairings you wouldn't have thought to try, and for generating varied batches from one graph.

The LoRA world rewards this kind of poking around. Interactions between stacked LoRAs are unpredictable - add one at a time is the usual advice precisely because you can't guess how two will blend - so a randomizer that shuffles combinations is a legitimate way to discover what works before you commit.

The workflow it fits: point it at a folder of style LoRAs, queue a batch of a dozen, and skim the results for the combinations that clicked. Then note the seed that produced a keeper - the config's seed makes any given draw reproducible, so a good accident isn't lost. It's exploration up front, reproducibility when you find something.

How it works

You hand it a set of LoRAs and a randomizer_config that governs the selection - things like how many to pick and the seed that makes a given roll reproducible. It draws from that set (or from a Lora Pool, if you connect one) and outputs a LORA_STACK containing the chosen LoRAs. That stack then goes into any stack-consuming loader, which does the actual model patching.

Inputs and outputs that matter

• randomizer_config (RANDOMIZER_CONFIG, required) - the node's configuration widget for the selection rules and seed.
• loras (LORAS, required) - the set of candidate LoRAs to draw from.
• pool_config (POOL_CONFIG, optional) - plug in a Lora Pool to draw from a reusable predefined pool instead.

One output: LORA_STACK - into a loader (or into Lora Stack Combiner to merge with a fixed stack).

How to install it

Via ComfyUI Manager: search lora-manager, Install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/willmiao/ComfyUI-Lora-Manager.git
cd ComfyUI-Lora-Manager
pip install -r requirements.txt

then restart. Light Python install, no models bundled.

Common issues & troubleshooting

The result is the same every run. The seed in the randomizer config is fixed. Randomize it if you want a new draw each generation - a static seed is what you'd use when you found a combo you like and want to lock it.

A random draw produced a broken image. Stacking LoRAs can trip black frames or NaNs, especially with strong LoRAs together. The community fix is to reorder with the larger LoRAs first; if a particular pairing keeps failing, prune it from the candidate set.

Randomizer vs Cycler? Randomizer picks at random; Lora Cycler steps through options in order. Use the Cycler when you want systematic, repeatable coverage (like sweeping every LoRA once); use this when you want surprise.