Nodes/ComfyUI_tinyterraNodes/pipeLoader v1 (Legacy)
ComfyUI Node Runs on cloud

pipeLoader v1 (Legacy)

The old loader that taught ComfyUI to use pipes

By TinyTerraΒ·Created 3 years agoΒ·Updated 3 months agoΒ· 604
pipeLoader v1 (Legacy)
  • model_override
  • clip_override
  • optional_lora_stack
  • pipe
  • model
  • positive
  • negative
  • latent
  • vae
  • clip
  • seed
β—„ckpt_nameβ–Ύβ–Ί
β—„config_nameDefaultβ–Ί
β—„vae_nameβ–Ύβ–Ί
β—„clip_skip-1β–Ί
β—„lora1_nameβ–Ύβ–Ί
β—„lora1_model_strength1.00β–Ί
β—„lora1_clip_strength1.00β–Ί
β—„lora2_nameβ–Ύβ–Ί
β—„lora2_model_strength1.00β–Ί
β—„lora2_clip_strength1.00β–Ί
β—„lora3_nameβ–Ύβ–Ί
β—„lora3_model_strength1.00β–Ί
β—„lora3_clip_strength1.00β–Ί
β—„positivePositiveβ–Ί
β—„positive_token_normalizationβ–Ύβ–Ί
β—„positive_weight_interpretationβ–Ύβ–Ί
β—„negativeNegativeβ–Ί
β—„negative_token_normalizationβ–Ύβ–Ί
β—„negative_weight_interpretationβ–Ύβ–Ί
β—„empty_latent_width512β–Ί
β—„empty_latent_height512β–Ί
β—„batch_size1β–Ί
β—„seed0β–Ί

pipeLoader v1 is the legacy loader that started the tinyterra pipe idea, and if you're loading old shared workflows, it's the node you'll keep bumping into. It loads a checkpoint, applies up to three LoRAs via dedicated slots, encodes positive and negative prompts with the advanced weight-interpretation options, builds the latent, and bundles it all into one PIPE_LINE output. It's flagged Legacy in the pack, and there's a version 2 - but the v1 is still around because so many published workflows use it.

The history is the interesting part: the README credits this as a modified mashup of Efficiency Nodes and BlenderNeko's ADV_CLIP_emb. It's the Efficiency Loader's "load everything in one box" philosophy with advanced prompt encoding bolted on and a pipe output added. If you came from A1111 and thought "why is my workflow 40 nodes," this was the pack that said the loader part doesn't have to be five nodes.

How it works

ckpt_name, config_name, vae_name, and clip_skip handle the checkpoint. Then the three LoRA slots: lora1_name / lora2_name / lora3_name, each with separate model_strength and clip_strength - the classic three-LoRA limit that v2's text-box approach removed. Prompts get the full treatment: positive_token_normalization and positive_weight_interpretation (comfy / A1111 / compel / comfy++ / down_weight), same for negative. empty_latent_width / height and batch_size build the latent - note v1 uses explicit width/height, not the aspect-ratio presets v2 added. seed is set here and carried in the pipe.

Optional inputs: model_override / clip_override to substitute models, and optional_lora_stack for a stack from another pack. Outputs: pipe plus the individual model, positive, negative, latent, vae, clip, seed.

The one setting that matters most

The positive_weight_interpretation dropdown is the v1's party piece: it lets you choose how (word:1.2) style weights are parsed. Most people pick comfy (native) or A1111 if migrating prompts, and compel if you want compel's richer syntax. Set it once per workflow and leave it.

Installing it

Part of tinyterraNodes:

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

or ComfyUI Manager β†’ tinyterraNodes. No extra dependencies.

Gotchas

You're installing a legacy node, so the honest advice is: if you're building a new workflow, reach for pipeLoader_v2 - same pipe output, text-box LoRAs (unlimited), aspect-ratio presets, prepend inputs, and ControlNet-stack support. But keep v1 around, because shared workflows are full of it and it still works fine. One real gotcha either way: the pipe type is shared between v1 and v2, so they interoperate - but the SDXL pipe (PIPE_LINE_SDXL) is a separate type and won't plug in here. If an old workflow loads with v1 and errors on load, it's usually a missing node warning from a different pack, not a pipe mismatch.

Category🌏 tinyterra/legacy

Inputs (26)

NameTypeDefaultDescription
ckpt_nameCOMBO0 options:
config_nameCOMBODefault12 options: Default, anything_v3.yaml, v1-inference.yaml, v1-inference_clip_skip_2.yaml, v1-inference_clip_skip_2_fp16.yaml, v1-inference_fp16.yaml, +6
vae_nameCOMBO1 options: Baked VAE
clip_skipINT-1-24–0β€”
lora1_nameCOMBO1 options: None
lora1_model_strengthFLOAT1.00-10–10β€”
lora1_clip_strengthFLOAT1.00-10–10β€”
lora2_nameCOMBO1 options: None
lora2_model_strengthFLOAT1.00-10–10β€”
lora2_clip_strengthFLOAT1.00-10–10β€”
lora3_nameCOMBO1 options: None
lora3_model_strengthFLOAT1.00-10–10β€”
lora3_clip_strengthFLOAT1.00-10–10β€”
positiveSTRINGPositiveβ€”
positive_token_normalizationCOMBO4 options: none, mean, length, length+mean
positive_weight_interpretationCOMBO5 options: comfy, A1111, compel, comfy++, down_weight
negativeSTRINGNegativeβ€”
negative_token_normalizationCOMBO4 options: none, mean, length, length+mean
negative_weight_interpretationCOMBO5 options: comfy, A1111, compel, comfy++, down_weight
empty_latent_widthINT51264–16384β€”
empty_latent_heightINT51264–16384β€”
batch_sizeINT11–64β€”
seedINT00–18446744073709550000β€”
model_overrideoptMODELβ€”
clip_overrideoptCLIPβ€”
optional_lora_stackoptLORA_STACKβ€”

Outputs (8)

NameTypeDescription
pipePIPE_LINEβ€”
modelMODELβ€”
positiveCONDITIONINGβ€”
negativeCONDITIONINGβ€”
latentLATENTβ€”
vaeVAEβ€”
clipCLIPβ€”
seedINTβ€”