tinyConditioning

tinyConditioning is the middle layer of the tinyterra "tiny" trio: tinyLoader hands you a raw model and CLIP, and this node turns them into ready-to-sample conditioning - positive and negative - while also applying any LoRAs you name, all in one box. It's the node you reach for when you want loader-style LoRA handling without the full pipe ceremony.

The idea is that your model and your prompt are separate concerns. tinyLoader (or any other model source) feeds this node; you type the prompt and the <lora:...> stack here; and out come positive and negative CONDITIONING plus the patched model, ready to hit a sampler. If you've ever applied a LoRA with three separate nodes and then had to re-encode the prompt by hand, this collapses the whole sequence.

How it works

It takes a model and clip, applies the loras text (a multiline box using the <lora:name:weight:optClipWeight> syntax the pack uses everywhere) to patch the model and CLIP, then encodes positive and negative through the pack's embedding encoder. Two things make it more than a bundled Apply LoRA + Text Encode:

• positive_token_normalization / weight_interpretation - the pair of dropdowns (comfy / A1111 / compel / comfy++ / down_weight) that control how prompt weights like (word:1.2) are parsed. Comfy's native interpretation is the default-minded choice; pick A1111 if you're migrating prompts over.
• prepend_positive / prepend_negative - optional wired STRING inputs for a prefix you keep separate, so a stable quality block can live in one place and get inserted ahead of the prompt. There's also zero_out_empty if you want an empty prompt to produce a zeroed conditioning instead of whatever the model defaults to.
• optional_lora_stack - accepts a LORA_STACK from another pack (like a multi-LoRA stack node) as an alternative to the text box.

Outputs are model (LoRA-patched), positive, negative, clip, and the pos_string / neg_string that echo the final composed text - handy for debugging or feeding a text output node.

Inputs you'll actually set

• positive / negative - your prompts, multiline, dynamic-prompt aware.
• loras - one per line, <lora:add_detail:0.8> style. This is the feature that makes the node worth it.
• The two interpretation dropdowns - set once per model family and forget.

Installing it

Part of tinyterraNodes:

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

or ComfyUI Manager → tinyterraNodes. No extra dependencies or model downloads.

Gotchas

The LoRA text syntax is the main trip point: <lora:name:weight> with colons, no spaces, and the model strength is what you're setting - clip strength is optional after a second colon. Get that wrong and the pack just ignores the line with a warning in the console, not an error you'll notice. Also note this node does not build an empty latent or batch size - that's tinyLoader's job - so wire them together rather than expecting one node to do everything. If you want the all-in-one with pipe throughput, the pipeLoader_v2 is the bigger sibling.