VELVET VICE MiniMax H3 — Optional Turbo LoRA

MiniMax H3 will render in plain "base" mode, but the pack author's advice is blunt: a compatible Turbo LoRA is strongly recommended because it can cut sampling cost dramatically. The VelvetViceMiniMaxH3OptionalTurboLora node is how that LoRA gets applied - and "Optional" is doing real work in the name, because when Turbo is off or the LoRA file is missing, this node passes the model through completely unchanged rather than erroring or silently skipping an expected file.

This loader is the companion to the Turbo master switch on the H3 Director. Flip Turbo on and this node looks for the official Turbo LoRA; flip it off and the node is a no-op that costs you nothing.

How it works

The node takes a model in, and its lora_name dropdown is set to AUTO - with the twist spelled out in the description: AUTO applies the official Turbo LoRA when it's present in any LoRA subfolder, and otherwise passes the model through untouched. The pack scans all your LoRA folders, not just the root, so a file sitting in a subfolder still gets found. A strength_model float (default 1.0) controls how hard the LoRA bites, and enabled (default false) is the on/off that makes "Turbo OFF remains truly optional."

The README gives you the matching vocabulary: when the console reports Turbo LoRA AVAILABLE/READY, the file was detected. When it's not, base mode just runs. And the LoRA is only applied when Turbo is actually enabled in the Director - the author is explicit that you should "match Turbo Steps to the LoRA you downloaded," because a turbo LoRA trained for a specific fast step count behaves differently at the wrong step count.

Inputs and outputs

• model (MODEL) - the diffusion model in.
• lora_name (AUTO) - the auto-detecting selector.
• strength_model (FLOAT, default 1.0) - LoRA blend strength.
• enabled (BOOLEAN, default false) - whether the turbo path is active.

Output: a single MODEL out. Straight pass-through when disabled.

Install and where the confusion starts

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/Velvet-Vice/velvet-vice-minimax-h3

or ComfyUI Manager → velvet-vice-minimax-h3, then restart. The LoRA itself is a model download you fetch separately - the author links MiniMax H3 Turbo LoRAs on Civitai from the README.

Where people get burned: they enable Turbo in the Director but never downloaded the LoRA, so the render runs at base steps and takes forever while the console quietly says the LoRA isn't there. The pack's philosophy is deliberately forgiving - missing files bypass back to base instead of halting - which means you have to notice the bypass. Read the status text at the turbo nodes before you queue a long render. And if you're stacking multiple H3 LoRAs at once, that's the Power LoRA AV node's job, not this one - this is the single dedicated turbo/distilled loader.