🚀 Isulion Mega Prompt V3

This is the node the pack is named after. The Isulion Mega Prompt V3 takes a theme dropdown - 82 of them, from Cyberpunk to Ghibli to "Peaky Blinders" - and spits out a complete, comma-joined positive prompt. If you've ever sat at 1 AM staring at a blank prompt box, this is the panic button.

It's not an AI. It doesn't call any API and needs no key. Under the hood it's a registry of per-theme Python handlers, each one randomly assembling subject, environment, style and effects from hand-curated word banks stored as JSON config files. Pick "Cyberpunk", and a CyberpunkThemeHandler grabs from lists like street samurai / netrunner / corpo assassin, then neon-lit alleyway / holographic skyline, then bolts on style and effects. Every theme in the dropdown has its own handler, which is why the pack ships with a theme_handlers/ directory holding ~90 files. The generator is a roulette wheel, not an LLM.

The inputs that matter

• theme - the whole point. Defaults to "🎲 Dynamic Random", which rerolls the theme every run. Set it to a specific one when you want Cyberpunk Tuesday to stay Cyberpunk.
• complexity - simple, detailed (default), or complex. This controls how many components get stacked into the final prompt.
• randomize - enable/disable. Disable to lock exactly what the dropdowns produce and use the seed instead.
• custom_subject / custom_location - your override levers. The README's whole pitch: type "Batman" into custom_subject and "Gotham" into custom_location, and every theme dresses your subject up instead of rolling its own. The handler splices them into the template while still adding theme flavor.
• lora_key - any text here gets appended to the end of the prompt. That's the intended slot for a LoRA trigger phrase, so you can chain a trained character/style in without hand-editing the output.
• include_environment / include_style / include_effects - kill switches. Set no to strip those sections if your checkpoint or LoRA already covers them.

Outputs: prompt (the full string - wire it into a CLIP Text Encode), plus selected_theme, subject, environment, style, effects broken out separately (handy if you want to feed parts into different conditionings), and seed so you can record what produced a given frame.

The ((word)) elephant in the room

The generated prompts lean hard on ((parenthetical emphasis)) syntax. That's SDXL-lineage language - it works great on SDXL, Illustrious, NoobAI, Pony and friends, where the CLIP encoder honors attention weights. On Flux and the newer LLM-encoded models those weights are silently discarded, so the output still reads fine as a sentence, it just loses its emphasis punch. The README's own examples are Flux workflows, so it does work there - just know the emphasis is doing nothing on those models. If you're on SDXL, leave the defaults and enjoy.

Install

ComfyUI Manager → search "Isulion" → install, or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/Isulion/ComfyUI_Isulion

Then restart ComfyUI. No requirements.txt, no model files, no heavy dependencies - it's pure Python word banks, so there's nothing to download and nothing to conflict with. The README's Ollama section only applies to the old chimera workflow that used a separate ollama_generator node; you don't need Ollama for this.

Gotchas

• Seed 0 + randomize enable means "pick a random seed" - the node generates one and hands it back on the seed output. Wire that output back into the seed input to lock a specific result.
• The random draws use Python's random, seeded per-run. Same seed, same theme, same word bank = same prompt.
• If the node ever reports "Theme handlers could not be loaded during startup", that's the theme registry failing to import one of its ~90 handler files (usually after a broken update). Reinstall the pack cleanly before debugging anything else.

It's a meat-and-potatoes prompt generation node that does exactly one thing well: turning a theme choice into a full prompt without you writing 200 words. Not a creative breakthrough, but it's saved more blank-prompt nights than I care to count.