DP Five Lora Random

Same five-LoRA-stack node as DP Five Lora, with one twist: the strengths randomize. Instead of pinning each LoRA to a fixed value, the random strength widget rolls fresh numbers, so every queued generation blends your five LoRAs a little differently. It's a variety-and-exploration tool - you're not hunting for one perfect setting, you're letting the machine sample the space between your LoRAs to see what combinations look good.

This is genuinely handy early in a project. You've got a handful of style LoRAs and no idea what mix works. Rather than manually grinding through strength combinations, load them all here, queue a batch, and browse the results. The winners tell you which LoRAs and roughly which strengths to lock in - at which point you'd switch to the fixed DP Five Lora for controlled output.

The inputs that matter

• model (MODEL) and clip (CLIP) in - your checkpoint's model and clip, passed through LoRA-modified.
• loader_state (ON / OFF) - master bypass. OFF sends the base model through clean, which is your reference point for judging what the random stack is actually doing.
• Lora_01 (required) and Lora_02–Lora_05 (optional) - the LoRA slots, dropdowns of your installed files. ("None" everywhere means your loras folder is empty.)
• strength_control (optional) - here it's the LORA_RANDOM_STRENGTHS widget, the piece that drives the randomization (setting the ranges each LoRA's strength rolls within).

Outputs: model, clip, and lora_info (a string logging what got loaded and - importantly with this node - the strengths that were actually rolled). Wire lora_info to a text preview so you can read back the winning combination when a random run nails it.

The workflow that makes it click

Explore with random, then commit with fixed. Queue a batch through this node, find the outputs you like, read their strengths off lora_info, then rebuild that exact mix in DP Five Lora for repeatable results. Randomization is for discovery; you don't ship on it.

How to install it

ComfyUI Manager: search ComfyUI-Desert-Pixel-Nodes, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/DesertPixelAi/ComfyUI-Desert-Pixel-Nodes

then restart. No download for the node; it lists LoRAs from ComfyUI/models/loras. Under the "DP" loaders category.

Common issues & troubleshooting

I found a great result but can't reproduce it. That's the trap with randomization - read lora_info for the rolled strengths and recreate them in the fixed DP Five Lora. Don't rely on re-rolling the same random result; capture the numbers instead.

Dropdowns show only "None". No LoRA files in ComfyUI/models/loras. Add some and refresh.

Results are all over the place, mostly bad. Random strengths across five LoRAs is a wide space, and a lot of it is ugly - that's normal. Narrow it: use fewer LoRAs, or tighten the strength ranges in the random widget so it rolls closer to values you already suspect work. And check every LoRA matches your base architecture; a mismatched one contributes noise no matter what strength it rolls.