WeiLin TwoInOne Prompt To AutoLoras

If you came to ComfyUI from A1111 or Forge, the first thing you miss is writing <lora:...> inline in your prompt and having it just work. Stock ComfyUI doesn't work that way - a LoRA is a separate loader node with its own strength dials. More power, sure, but a lot more clicking. This node is the bridge: type the tag, and it parses it, loads the LoRA, patches your model and CLIP, encodes both prompts, and hands you ready-to-use conditioning.

What it replaces

One node doing the job of the whole Load Checkpoint → LoraLoader → CLIPTextEncode (pos) → CLIPTextEncode (neg) chain. You give it a model and CLIP from your checkpoint loader, type both prompts, and out come a patched model and two CONDITIONING outputs that wire straight into a KSampler.

How it works

On every run it scans the positive prompt for <lora:...> tags, pulls each one out (along with any trailing comma), and for each tag:

1. Takes the filename inside the tag and resolves it against your ComfyUI/models/loras folder.
2. Loads the .safetensors and patches both the model and CLIP using ComfyUI's own LoRA machinery.
3. Encodes the cleaned-up prompt text into a proper CONDITIONING.

So the tags never survive into the conditioning text - the prompt you actually condition on is the prompt minus the lora tags.

The strength syntax (straight from the pack README):

• <lora:filename.safetensors:0.3:0.4> - 0.3 model strength, 0.4 CLIP strength
• <lora:filename.safetensors:0.4> - both set to 0.4
• <lora:filename.safetensors> - both default to 1.0

Two practical notes. The filename has to match what's in your loras folder, extension included - the editor inserts it that way, so copying a tag you wrote by hand can silently fail if you dropped the extension. And the same architecture-bound rule as always applies: a LoRA only works if it's built for the checkpoint you loaded. An SDXL LoRA on a Flux model does nothing.

Inputs and outputs

• model (MODEL) and clip (CLIP) - from your checkpoint or model loader.
• positive and negative (multiline strings).
• Outputs: model, 正向条件 CONDITIONING, 负面条件 CONDITIONING (yes, the output names display in Chinese for everyone - that's just what the author named them).

Wire model and both conditions into a KSampler, and that's the whole workflow.

The catch about negatives

The negative prompt does not parse LoRA tags. The code encodes the negative as-is - a <lora:...> tag in the negative box stays literal text. In practice you rarely want a LoRA in the negative anyway, so this is more "know it's not there" than a real limitation.

Installing

The easy way: ComfyUI Manager → Custom Nodes Manager → search WeiLin-ComfyUI-prompt-all-in-one → Install → restart ComfyUI. The manual way:

cd ComfyUI/custom_nodes
git clone https://github.com/weilin9999/WeiLin-ComfyUI-prompt-all-in-one

On first launch it pip-installs its two Python deps (gradio and ruamel.yaml) automatically, so give the console a minute before you panic.

Should you bother?

Honest take: for serious multi-LoRA work, rgthree's Power Lora Loader is the better tool - per-LoRA toggles, trigger words pulled off the file or from CivitAI, no tag-parsing to trip over. But this node is the only one of the five in this pack that does something stock ComfyUI can't, and it shines when you copy a LoRA prompt off a model page or C站 and just want it to run. One big caveat to carry in: the pack is deprecated. The README's first line says it's no longer accepting feedback or updates, and the successor (WeiLin-Comfyui-Tools) isn't node-compatible, though your tag data can migrate. Works fine today; don't plan a decade of workflows on it.