🎥 Isulion Video Prompt Generator

Most nodes in the Isulion pack hand you a single word. The Video Prompt Generator is the exception: it composes a complete, sentence-length video prompt - subject, location, camera movement, lighting, and a visual style - in one go. It's the biggest node in the pack and the one that most resembles what a text-to-video model (Wan, Hunyuan, or the API crowd) actually wants to read.

Think of it as Mega Prompt V3 in video clothes. It shares the mega generator's theme engine - all 83 of those themed prompt handlers - and wraps them in the vocabulary video models care about: camera angles and lighting conditions.

How it works

It reads a JSON config (video_prompt_config.json bundled in the pack) that holds subject and location pools, camera-angle phrases, lighting phrases, quality modifiers, and the preset scenarios. When you pick a preset like "Nature Documentary," it pulls the preset text, appends your custom subject/location if you gave them, then tacks on a camera move, a lighting condition, and a style from the theme list. No preset? It rolls a random subject and location from the config pools instead and does the same assembly. Everything is seeded, and the output ends with a period - it's built to read like an instruction, not a tag list.

The inputs that matter

Nine inputs, and you'll mostly touch four:

• preset - 11 scenarios: Nature Documentary, Travel Vlog, Wedding Video, Music Video, News Report, Sports Coverage, Fashion Show, Food Review, Real Estate Tour, Product Commercial. This is the scene skeleton.
• camera_angle - 17 options including "Random" and specific moves like "The camera pans slowly" and "The camera circles around." Camera moves are the highest-leverage text in video prompts, so this is the most useful input here. (The options ship pre-wrapped in (( )) - more on that below.)
• lighting - 22 conditions, from "golden hour sunlight bathes the scene" to "moody low-key lighting."
• theme - 83 themes inherited from the Mega Prompt V3 engine: Anime, Cyberpunk, Ghibli, Steampunk, Film Noir, and the rest. This is where the visual style comes from.
• custom_subject / custom_location - override the random subject/location. Empty means the node rolls its own from the config pools.
• quality_level - Standard, High Quality, Best Quality, Experimental, Raw Footage, Film, Broadcast. Prepended as a quality phrase.
• seed - seeds the whole composition. Seed 0 = new roll every run.

Output

A single STRING - the assembled video prompt. Wire it into your text-to-video model's prompt input. If you're running Wan or Hunyuan locally, this replaces the "write a sentence describing a moving scene" manual work.

Installing it

One pack, all Isulion nodes:

1. ComfyUI Manager → Custom Nodes Manager → search Isulion → Install → Restart.
2. Or clone it manually:

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

No requirements.txt, no model downloads. The only dependency is the mega-prompt engine, which ships in the same pack.

Common issues

The (( )) wrapping is the thing to know. Camera-angle phrases come out wrapped in double parentheses - classic CLIP-era attention-weight syntax. On a CLIP-based video pipeline that boosts emphasis; on an LLM-encoded video model (Wan's text encoder, most 2025+ models) those parentheses are inert and can even look odd if the model was trained on clean captions. If your output feels literal about the brackets, strip them. Second, this node is heavy on jargon that video models may not all honor - "The camera follows the subject" works on some backends and gets ignored on others; treat camera moves as a suggestion you may need to phrase for your specific model. Third, it's a compositor, not a magician: the quality of the result tracks the quality of your base model, and preset texts are generic. Still, as a starting point for batch-generating varied video prompt ideas, it's genuinely handy - roll a dozen, keep the good ones.