Loader (Mango + 10 Loras)

This is Loader (Mango) with the knob turned to eleven - literally. Same checkpoint dropdown, same model/clip/vae outputs, same metadata plumbing, but ten LoRA slots instead of five. If your workflow genuinely stacks more than five LoRAs - a style, a character, a couple of detail enhancers, a lighting mod, and a few odds and ends - this node saves you from chaining three loaders in series.

The honest question to ask yourself before you reach for it: do you actually need ten LoRAs, or do you need to prune your stack? Stacking ten LoRAs at weight 1.0 each is a recipe for mush, overlapping styles fighting each other, and every prompt tag competing for attention. This node is for the edge cases where you truly have a big, deliberate stack - not a license to skip curation.

How it works

Mechanically it's identical to Loader (Mango): it loads the checkpoint with comfy.sd.load_checkpoint_guess_config, computes a 10-character SHA-256 hash of the file, applies each selected LoRA with comfy.sd.load_lora_for_models in slot order, and records the applied LoRAs as a lora_stack of (name, weight, weight) tuples. The only difference is the loop runs ten times instead of five. LoRAs apply in slot order, so slot 1 goes in first and later slots get applied on top.

Inputs and outputs

• ckpt_name - your checkpoint dropdown.
• LoraName1…LoraName10 / LoraWeight1…LoraWeight10 - ten LoRA slots, weights −10 to 10, default 1.0. Any slot set to None is skipped.

Outputs are the standard Mango loader set: model / clip / vae, plus lora_stack, ckpt_name (STRING), and ckpt_hash (STRING). Feed the stack and the ckpt strings into a Mango sampler so your saved metadata records the full ten-LoRA picture.

Installing it

It's part of the Mango Node Pack:

cd ComfyUI/custom_nodes
git clone https://github.com/mang01010/MangoNodePack

Restart ComfyUI, or install "Mango Node Pack" from ComfyUI Manager. The README's only extra dependency is pip install safetensors. No extra models needed - it reads whatever checkpoints and LoRAs are already in your models folders.

Gotchas

Performance is the real trade-off. Every LoRA you add is another weight merge into the model and CLIP, and ten merges add up - expect a slower load than the five-slot version, especially on big SDXL-era checkpoints. There's also the Civitai hash recognition issue the pack README warns about: occasionally a LoRA's hash won't match anything on the site because the model was updated after you downloaded it - re-downloading fixes it. And while ten slots sound great, remember that each one is applied at whatever weight you set; five strong, well-chosen LoRAs will almost always beat ten at half strength. Use the extra headroom for the two or three niche LoRAs you occasionally want, not as a default.