Bonus Prompt Video Presets

Video prompting is image prompting with extra failure modes - and if you've tried to hold a style across 100 frames of MiniMax H3 or Wan output, you know the style block is doing a lot of the work. This node is the video-tuned sibling of UC_BonusPromptPresets: same idea, a dropdown of ready-made style paragraphs, but written for the way video models actually want to be prompted.

What's different? The presets in the video variant lean into motion language - describing the style as something that applies to a scene over time, with phrasing that anchors lighting, camera, and material behavior rather than just "this is a render." That's the right instinct, because video models tend to collapse styles toward their training prior unless the instruction is repeated, explicit, and motion-aware. A still-frame style sentence like "This is a 3D render" carries less weight when the model is also deciding where the camera goes and how things move - the video presets front-load the aesthetic so it survives the motion.

The node itself is identical in shape to its image sibling: a preset dropdown (default STYLE_3D), an escape_parentheses boolean whose tooltip is blunt about its job - "prevents errors from parentheses left in the prompt" - and one bonus_prompt string output. Wire it into whatever prompt-combining node feeds your video encoder's positive conditioning. There's no model involved, no sampling, no VRAM cost; it's pure text.

Where people get tripped up is the same place they trip on any preset node: they treat the dropdown as a complete prompt instead of a style modifier. Your actual instruction - "the camera slowly orbits a ruined cathedral at dusk" - is the part doing the narrative work. The bonus block is the aesthetic frame around it. Concatenate them (bonus block first, your action prompt second, is a reasonable default for the Qwen-family encoders), don't stack multiple style presets, and you'll get the consistency without burying the actual scene description.

Worth knowing for the wider pack: this preset family - image, video, camera shot, edit target, edit operation - exists because the author ships it alongside their MiniMax H3 and VLM encoder work, where prompt structure genuinely changes output. These aren't a substitute for understanding your model, but they're a better baseline than retyping "cinematic, 8k, masterpiece" in 2026.

Install is the pack standard: ComfyUI Manager search "ComfyUI-UtilsCollection", or git clone https://github.com/silveroxides/ComfyUI-UtilsCollection into ComfyUI/custom_nodes and restart. No model downloads, no dependencies beyond the pack's opencv-python and typing-extensions. If you already have the pack installed for the MiniMax H3 nodes, this one was already on your node list - it just lives a few taps away in the menu.