pipeLoader

pipeLoader is the heart of the tinyterraNodes pack: one node that loads your checkpoint, applies LoRAs, encodes both prompts, builds the empty latent, and bundles model + conditioning + latent + VAE + CLIP + seed into a single PIPE_LINE output. The pipe travels down one wire into pipeKSampler, and that's what collapses an eight-node workflow into two boxes.

This is the "everything but the sampler" node. It borrows the idea from Efficiency Nodes - a pack that's explicitly cited in the README as the ancestor - and adds a couple of things that made it a fixture in shared workflows: an embedded Advanced CLIP Text Encode (from BlenderNeko's ADV_CLIP_emb, which it also credits) with token-normalization and weight-interpretation dropdowns, and a LoRA text box instead of one-per-slot dropdowns.

How it works

The ckpt_name, vae_name, and clip_skip handle the checkpoint side. The loras box is the standout: a multiline field where each line is <lora:name:weight:optClipWeight>, so you can stack as many LoRAs as you want without adding nodes - the pack's autocomplete even suggests installed LoRAs as you type. empty_latent_aspect gives you the same resolution presets as tinyLoader (plus custom width/height), and batch_size sets the latent batch.

The prompt side gets the advanced treatment: positive / negative with the four token-normalization options and five weight-interpretation modes (comfy, A1111, compel, comfy++, down_weight). Optional prepend_positive / prepend_negative let you wire in a stable prefix block, and optional_lora_stack / optional_controlnet_stack accept stacks from other node packs.

Inputs you'll actually set

• ckpt_name and vae_name (Baked VAE unless your model needs a separate one)
• loras - one <lora:...> per line
• positive / negative - your prompts
• seed - set it here; the sampler will pick it up from the pipe

Outputs: the pipe plus individual model, positive, negative, latent, vae, clip, seed, and - usefully - width, height, pos_string, neg_string. You can grab any of these directly instead of digging into the pipe.

Installing it

Part of tinyterraNodes:

cd ComfyUI/custom_nodes
git clone https://github.com/TinyTerra/ComfyUI_tinyterraNodes.git

or ComfyUI Manager → tinyterraNodes. No pip dependencies, no model downloads beyond your own checkpoints/LoRAs.

Gotchas

The pipe format is the pack's own PIPE_LINE, so a pipeLoader_v2 wire only plugs into nodes that speak it - pipeKSampler_v2, pipeEDIT, the pipe→basic_pipe converter for Impact Pack compatibility, and so on. The legacy v1 pipeLoader uses the same pipe type, so v1 and v2 mix; the SDXL variant uses a different PIPE_LINE_SDXL, which is the one people hit when they cross-wire a normal pipe into an SDXL sampler and nothing happens. Also note the LoRA autocomplete needs the feature enabled in the pack config (it's on by default), and clip_skip defaulting to -1 will quietly change results on models that don't want it.