Local Lora Gallery Stack Apply

This is the working half of the pack's split personality. Where the Stacker builds a LORA_STACK and applies nothing, the Local Lora Gallery Stack Apply takes a stack in and gives you a loaded MODEL (and CLIP, if you want one) out. It's the plain-vanilla application node - and because it accepts a LORA_STACK from anywhere, the gallery doesn't have to be in the same graph as your sampler.

Why you'd reach for it

If you're using the Stacker for its clean separation - gallery as data, apply later - this is the natural partner: lora_stack from the Stacker into this node, model from your checkpoint loader, and the outputs go where the LoraLoader normally would. But it's not limited to the pack's own gallery. Any node that produces a LORA_STACK can feed it, which makes it a handy generic LoRA-stack applier even if you never look at a gallery card.

The interesting detail is the optional clip input. Leave it unhooked and the node applies LoRAs model-only. Connect it and it applies both halves. That's a nicer shape than ComfyUI's default, where you're either on the model-only loader or the full one and switching means swapping nodes.

How it works

apply_lora_stack walks the incoming stack in order - order matters, since stacked LoRAs interact - and pushes each entry through ComfyUI's own LoraLoader with its model and clip strengths. If clip isn't connected, it falls back to LoraLoaderModelOnly so the graph still runs. Like the rest of the pack, it auto-detects Nunchaku-accelerated models and swaps in the accelerated loader when the model is a Nunchaku wrapper (Flux, Qwen, Z-Image), falling back to standard loading otherwise.

Inputs and outputs

Required: lora_stack (LORA_STACK) and model (MODEL). Optional: clip (CLIP). Outputs:

• MODEL - into your sampler (via whatever conditioning chain you use).
• CLIP - only meaningful if you connected clip; wire it to your text encoder.

An empty stack is a no-op - it passes the model and clip straight through, which is handy for workflows where the stack might legitimately be empty.

Installing it

Same pack, same story - no Python dependencies beyond ComfyUI's built-in loaders.

cd ComfyUI/custom_nodes
git clone https://github.com/Firetheft/ComfyUI_Local_Lora_Gallery.git

Restart ComfyUI, or install via ComfyUI Manager by searching "Local Lora Gallery". The optional comfyui-nunchaku plugin only matters if you run accelerated models.

Gotchas

• Empty stack, no error, no change. If nothing comes out the other end different, check that your stack source actually produced entries.
• Strength lives in the stack, not on this node. There are no strength knobs here - whatever values the Stacker (or your other stack producer) baked into the tuples are what get applied. Edit the source.
• Clip is optional for a reason. If a stack was built with a clip strength in mind and you leave clip unhooked, that half is silently dropped. For model-only bases like most Flux work, that's exactly right.

If you like your graph tidy - gallery on one side, apply node near the sampler - this completes the picture. And if you already own stack-producing nodes, it's a drop-in way to apply them without fighting ComfyUI's single-LoRA loader.