Lazy-subject-and-scene-automation

This is the flagship node of the pack, and the one worth building a workflow around. It takes one subject file, one or two scenario files, and a pile of LoRA slots, and assembles everything into a ready-to-encode prompt plus properly LoRA-stacked model/CLIP outputs. Where most people wire up a chain of subject selector → scenario selector → several LoRA loaders → text concat nodes, this node does the whole middle of the graph in one box.

The context that makes it click: Wan-style video graphs run two model branches - a high-noise and a low-noise stack - and every LoRA has to be applied to both, usually with different strengths. That's why a single Wan workflow ends up covered in LoRA loaders. This node models that directly: wire model_high and clip_high (required), and for a dual-stack Wan graph also wire model_low and clip_low. Running single-model (Z-Image, Krea2, one Flux/SDXL branch)? Just leave the low inputs unwired and use the high outputs - the low branch is skipped.

How it works

Your subject and scenario definitions live in plain .txt files under the node's own folders:

custom_nodes/vsaan212_workflow_utilities/lazy_subject_scene_automation/
├─ SubjectFiles/
└─ ScenarioFiles/

The files use a simple tagged format. A line like [LoraHighA][1.0][0.7] opens a LoRA slot with model and clip strengths; the next lines are the LoRA path (or the word bypass). [desciption] holds the text that goes into your prompt, [KeywordA] feeds the keywords output, and [Prompt] is special - it's routed out on prompt_override for an LLM to expand, not dumped into the prompt. Subject files get slots A/B/C; each scenario file gets another A/B/C, so with scenario_2 you can stack up to six scenario LoRA sets. {left|right} inside scenario text picks a random option on every queue - nice for batch variety without an external random-prompt node.

LoRAs apply in order: subject slots, then scenario 1, then scenario 2, using ComfyUI's stock LoraLoader. A missing or bypass slot is skipped. And there's a Wan 2.1-style fallback: if a low slot is bypassed but its high twin has a real path, the same LoRA gets applied to both branches with clip strength forced to 1.0 on the low side.

Inputs and outputs that matter

The dropdowns - subject, scenario, scenario_2 - are where you pick your .txt files (relative paths, no extension). scenario_2 defaults to none; when it's set, the scenario_2_high_strength / scenario_2_low_strength sliders let you override that file's [LoraHighA] / [LoraLowA] model strength live. prepend_text and post_text are STRING sockets for framing text, and pass_subject_to_main_prompt controls whether the subject description lands in the main prompt.

The best part is the live editor. The pack ships a browser extension that adds editable text panes right on the node - subject, scenario, scenario 2. Queue uses the pane text, not the file on disk, and a Save edits button writes your changes back to .txt. Edit in the graph, persist when it's good.

Outputs: prompt (the assembled text for CLIP/preview), model_high / model_low and clip_high / clip_low (after the LoRA stacks - these feed your sampler), keywords, subject_description (wire it to LazyPrompt's character), and prompt_override (wire to LazyPrompt's prompt_override_input when a scenario file uses [Prompt]).

Install and gotchas

Same pack as everything else: ComfyUI Manager → search vsaan212/Vsaan212-workflow-utilities, or clone into custom_nodes and restart. Note the folder: this node reads its own lazy_subject_scene_automation/SubjectFiles, not the folders the standalone Subject/Scenario Selectors use - copy or symlink if you want the same files in both.

New .txt not in the dropdown? Press R in ComfyUI or recreate the node. Same filename in two folders? Use the full relative path in the dropdown. For a no-LoRA run, pick none (or set slots to bypass) - the pack seeds none.txt and a Bypass and format example.txt on first load so you always have working examples.