🔄 Isulion Multiple Prompt Generator

If the Mega Prompt V3 is the pack's single-shot, this is the burst-fire version. The Isulion Multiple Prompt Generator takes the same ~90 theme handlers and runs them all at once, returning a list of prompts instead of one string. Type a subject, pick the themes you care about, and out comes a ready-made batch for generating variations.

This is the node you want when you're doing a style comparison - the "what does my character look like as Ghibli vs. Cyberpunk vs. 1800s photography" workflow. It's built around the same theme handlers as the Mega Prompt V3, so the mechanism is identical: per-theme word banks, random assembly, no API, no LLM. The difference is plumbing. It returns lists, and your sampler needs to consume lists.

How the inputs work

The required set is tiny:

• selection_mode - Theme Selection or All Themes. "All Themes" is the scattergun: every theme, every run. Useful for idea mining, wasteful if you only care about five.
• custom_subject / custom_location - the same overrides as the Mega Prompt node. They get spliced into every theme's template, so one "Batman in Gotham" entry produces the whole menu.
• seed - the master seed for the batch.

Then come the 90+ optional booleans, and this is where it gets slightly unwieldy. There are 14 category toggles (category_science_fiction_fantasy, category_holiday_themes, …) and below them, per-theme checkboxes (🎌_anime, 🌆_cyberpunk, 🎄_christmas, …). Check the categories you want to include, or drill into individual themes. There's also a theme_names enum for directly choosing themes - but honestly, the checkbox grid is the intended UI. With 90 booleans you will spend your first session just finding the ones you want; my advice is to decide on 5–10 themes and check only those, because every checked theme is another prompt in the output.

Outputs: it's lists, not strings

• positive - a STRING list of the generated prompts, one per theme.
• name - a STRING list of the theme display names, in the same order. Never trust your memory on which index is which; wire name into whatever labels your batch.

Wiring this into a sampler is the part that trips people up. A plain KSampler's positive input wants a single conditioning, and you can't just drop a list in. The pattern that works: feed the list through a batch-capable text encoder (a "Text Encode with batch" style node) or pair it with a loop/variation node, then run the batch. In the stock-UI path you're better off using one prompt at a time via an index or a batch splitter. The README's example workflow shows it chained into a full batch setup, so if you're new to batched generation, start there.

Install and quirks

Same as every node in this pack: ComfyUI Manager → search "Isulion", or git clone https://github.com/Isulion/ComfyUI_Isulion into custom_nodes/ and restart. Zero extra dependencies, zero model files.

Two things to know before you blame it for broken outputs. First, it inherits the Mega Prompt's ((word)) emphasis syntax, which is live on SDXL-lineage models and inert on Flux-class encoders - the words still render, the emphasis just doesn't push. Second, seed 0 with randomize effectively on means every theme re-rolls its picks each run; if you want a reproducible batch, set an explicit seed. And remember the outputs are lists - if your downstream node is choking on a "STRING" that's actually 15 strings, that's expected, not a bug.