MiniMax H3 Optional LTX Detailer LoRA

LTX is the detailer in the H3 Shotboard - the model that upscales and sharpens the native render. Detailers often get a LoRA nudge, and this node is how the pack lets you add exactly one, by name, without hardcoding filenames into the graph. It's small, it's optional (default off), and its only job is "if enabled and a LoRA is named, apply it to the model; otherwise pass the model through untouched."

How it works

The mechanism is a single stock call: LoraLoaderModelOnly with your strength, applied to the incoming model. The twist is the escape hatches. If enabled is off, the lora_name is empty, or the strength is 0.0, the node returns the model unchanged with the report "LTX detailer LoRA off" - no error, no wasted load. If the named LoRA can't be found in models/loras, it logs a warning and falls back to the base LTX path rather than crashing. That fallback behavior is deliberate: an optional finishing nicety should never take down a multi-hour render.

There's one compatibility nicety buried in the code: if the LoRA filename contains "ic-lora-detailer" (the IC-LTX detailer LoRAs), the report adds a note that it's a compatibility load only - the IC Detailer's full effect needs the dedicated IC conditioning/guiding-latent LTX pipeline, which the standard Shotboard path doesn't run. Useful honesty, and the kind of thing you'd otherwise only learn after a wasted render.

Inputs and outputs

Inputs: model (the LTX diffusion model), enabled (BOOLEAN, default false), lora_name (a plain STRING - you type or paste the filename, it doesn't scan your folders), and strength (FLOAT, default 0.6, range 0–2). Outputs: the patched model and a report string telling you whether it applied, skipped, or fell back.

Installing it

Part of IAMCCS-nodes - ComfyUI Manager (search "IAMCCS") or git clone https://github.com/IAMCCS/IAMCCS-nodes.git into custom_nodes, restart. The LoRA itself you drop into models/loras; no extra packs required.

Where people get burned

The lora_name is a raw string, not a dropdown, so a typo silently produces the "unavailable; base LTX path retained" fallback - and a workflow that looks like it has a detailer but doesn't. If your upscale quality mysteriously doesn't change when you toggle this, read the report output: it'll say "off", "applied @strength", or "unavailable". Also remember the model-only application: this patches the diffusion weights, which is what the standard H3→LTX path wants. Full IC-guiding workflows should use the dedicated pipeline instead of expecting this node to replicate it.