Anima Prompt Studio Advanced LoRA

If you've come from A1111, you know the feeling: you type <lora:name:0.8> in the prompt and the LoRA just... loads. ComfyUI never worked that way - LoRAs are graph nodes, not prompt syntax. Anima Prompt Studio Advanced LoRA splits the difference: it's the full Advanced editor, plus it scans your prompt fields for <lora:...> tags and emits an actual LORA_STACK for the graph to consume. The prompt stays human-readable, the LoRA stack still ends up on the model.

What it is

Same editor and execution lifecycle as Prompt Studio Advanced - reorderable positive/negative fields, NAIA fill, resolution buckets, wildcard control, Mod Guidance routing, metadata outputs. On top of that it adds one optional input and one extra output:

• lora_stack (optional) - a stack you feed in from Anima LoRA Preset or another source.
• LORA_STACK (output) - your input stack, followed by every LoRA extracted from enabled positive fields, in field order.

So the output stack is "what you gave me, plus whatever <lora:...> tags appeared in the prompt." Wire it to a LoRA loader or straight into the AiO Generator's lora_stack socket and the tags in your prompt become applied LoRAs.

The syntax it understands

The extraction targets A1111 and LoraManager style:

<lora:name:model[:clip]>

name is the LoRA file, model is the model-side weight, and the optional clip weight is the text-encoder side. Two LoRAs in a field produce two entries in the stack. Tags are extracted from enabled positive fields in the order the fields appear, so prompt order maps to stack order - put your strongest LoRA earlier if ordering matters to you.

The extra output is what makes this node worth it over the plain Advanced version: one wire to the model and your prompt is its own LoRA manifest. Combined with the pack's wildcard support (a wildcard can expand into a <lora:...> tag at queue time), it's the full "prompt embeds its own LoRAs" workflow in one node.

The inputs that matter

Everything meaningful is inherited from Prompt Studio Advanced: use_naia / consume_naia_on_queue for NAIA fill, use_anima_mod_guidance / use_negative_anima_mod_guidance for quality routing, and the resolution/wildcard controls. The internal advanced_fields JSON is front-end managed - don't hand-edit. Outputs are the standard ten (positive, negative, mod guidance x2, booleans x2, metadata x2, width, height) plus the LORA_STACK.

Install and gotchas

Same pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/n0va39/ComfyUI-EasyUseAnima
cd ComfyUI-EasyUseAnima && pip install -r requirements.txt

Restart ComfyUI; Manager: "ComfyUI EasyUse Anima".

Where people trip: a malformed tag (missing colon, wrong filename) usually produces no stack entry rather than an error - check your LORA_STACK output with a display node if nothing seems to apply. And the extraction only reads enabled positive fields; a <lora:...> sitting in the negative field or a disabled field is silently ignored. LoRA weights are the A1111 convention (0.5–1.0 typical); Anima LoRAs generally train fast and stack well, but three or more at high weight is where even this model starts to bleed.