FireRed LoRA Loader (Trent)

The whole Qwen-Edit LoRA library - AnyPose, Multiple-Angles, Clothing Transfer, and the rest - mostly works here, and "mostly" is doing more work than it looks like. Because FireRed-Image-Edit-1.0 is a very shallow finetune of Qwen-Image-Edit-2509, the LoRA ecosystem built for 2509 is effectively the LoRA ecosystem for this model. FireRed LoRA Loader (Trent) is the node that lets you feed those files into the loaded pipeline, and it's the second node in the pack - the one that makes the editor do posing, camera angles, and garment swaps instead of just plain edits.

What it does

It takes the firered_pipe from the FireRed Loader, picks a LoRA from your ComfyUI loras folder, and applies it via diffusers' load_lora_weights + set_adapters calls (which is why peft is in the pack's requirements). The output is the same firered_pipe type, so you can chain several of these in a row and stack LoRAs - a base style plus a pose LoRA at different strengths, say.

Three inputs, and only one of them is a judgment call:

• firered_pipe - in from the Loader (or a previous LoRA loader).
• lora_name - a dropdown of every .safetensors file in ComfyUI/models/loras.
• strength - a float from -2.0 to 2.0, default 1.0. Negative strengths are allowed, which is handy for actively pushing against a LoRA's effect rather than just turning it down.

That's it - one output, the pass-through pipe.

The gotcha that will confuse you

This node mutates the cached pipeline in place. The Loader holds one pipeline in memory, and the LoRA loader edits that same object - so the LoRA stays applied even if you later delete the LoRA node, and its effect persists across runs until you change the Loader's settings (which triggers a full reload) or run the Unloader. If your next generation still looks "posed" and you swear you removed the node, that's why. It's not a bug, it's the price of the cache the Loader advertises.

Ground rules from the ecosystem

Because FireRed ≈ 2509, community Qwen-Edit LoRAs trained on 2509 (or 2511, which accepts them) generally load without drama. One honest caveat: this node loads the full diffusers pipeline, so the LoRA file must be diffusers/PEFT-compatible - the standard safetensors LoRAs from the Qwen-Edit library are, and that's what you'll find on HuggingFace. Strength 1.0 is a sane starting point; the pose and angle LoRAs often sit around 0.7 in the community workflows, so don't be shy about dialing down.

Install is the same as the rest of the pack - ComfyUI Manager, search "ComfyUI-FireRedTrent", or git clone https://github.com/TrentHunter82/ComfyUI-FireRedTrent into custom_nodes and restart. No model downloads of its own; the LoRAs are yours.