Sequential LoRA Loader

Sequential LoRA Loader is the one sequential loader in the u5 FreeVRAM pack that takes real inputs, because a LoRA has nothing to apply to by itself. It's ComfyUI's standard LoraLoader with a trigger wired in: it waits until told, then applies the LoRA to whatever model and CLIP you feed it.

The use case is the two-pass workflow the pack is built around. Your rough pass is a quick sketch to establish composition - you don't want the character LoRA and the style LoRA chewing up VRAM and time on that. Your finish pass is where the LoRAs matter. So you load the checkpoint, apply the LoRAs only after the FREE VRAM node cleared the rough pass's models, and sample the good stuff. That's this node's entire job.

How it works

It's a thin wrapper around ComfyUI's own LoraLoader: it takes the incoming MODEL and CLIP, applies the LoRA at the given strengths, and returns both, plus the relayed trigger. Same math, same outputs as the core node - the only difference is the ordering control.

Inputs and outputs

• model, clip - wire these from whatever loaded the model, typically a Sequential Checkpoint Loader in the same second pass.
• lora_name - dropdown of your models/loras.
• strength_model and strength_clip - both default 1.0, range -20 to 20 (negatives invert the effect, standard LoRA behavior). Leave at 1.0 unless you know why you're changing it.
• trigger (optional, any type) - the ordering control; relayed through.
• Outputs: MODEL, CLIP (both with the LoRA applied) and Bypass(Relay).

To stack multiple LoRAs, chain them - each Sequential LoRA Loader takes the previous one's MODEL/CLIP outputs, and you wire each trigger through the Bypass(Relay) outputs so ordering holds:

[Seq Checkpoint Loader] → [Seq LoRA (style)] → [Seq LoRA (character)] → [KSampler]
                              └─trigger──────────┘

Installing it

Comes with the u5 FreeVRAM pack. ComfyUI Manager → search "u5 FreeVRAM" → Install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/u5dev/comfyUI_u5_VramFREE
# restart ComfyUI

No model downloads, no new dependencies - it reuses ComfyUI's own LoRA loader.

Where people get burned

• Wire every trigger. The ordering guarantee is per-node, so if you chain three of these and forget one trigger link, that loader runs whenever it wants and your carefully sequenced second pass becomes just a normal workflow with extra steps.
• Stacking order and strength are on you. LoRAs apply multiplicatively in the order you chain them, and the community's first advice when stacking goes wrong is to reorder (larger/stronger LoRAs first) and recheck strengths. The node doesn't police any of that - it just applies what you give it.
• It doesn't free anything itself. It's the sequencing that saves VRAM. The FREE VRAM node in front of the chain is what actually unloads the first pass's models; skip that and this loader is just a LoraLoader wearing a costume.