Bonus Prompt Presets

If you've ever stared at a prompt block that says "This is a professional photograph with real life detailed textures and real life natural lighting" and thought "someone typed that by hand," this is the node that node came from. UC_BonusPromptPresets is a dropdown full of ready-made style paragraphs - the pack author calls them bonus prompts - that you append to your own prompt to push the whole generation toward a consistent style.

The presets are long, structured, instruction-style paragraphs, which is exactly the shape that works on 2026's LLM-encoded models. The KB's prompt-engineering doc makes the point well: your prompt is now an instruction, not a token bag, and models like Krea 2 and the Qwen-encoder family respond to blocks like "Modify any subjects' appearance to show rendered details. Make it look like a high-end CGI render." That's not the old masterpiece, best quality noise - that died with the encoder swap. These are full sentences with intent, written for the way current text encoders actually read.

The list is genuinely long: STYLE_3D (the default), PHOTOGRAPHY_PHOTOREALISM, ANIME, DIGITAL_PAINTING, PIXAR, GHIBLI, CYBERPUNK, COMIC_BOOK, film-emulation photography styles (Kodachrome, Polaroid, analog film), game styles (Mario, Minecraft, Skyrim, PS1), cartoon styles (Rick and Morty, South Park, Adventure Time)... the author has clearly been collecting these across their Chroma and Krea workflows. You get a preset dropdown, one boolean (escape_parentheses), and a single bonus_prompt string out. escape_parentheses is the one safety valve you'll actually use: if a preset contains stray parentheses, flipping it on escapes them so a strict text encoder doesn't try to parse them as weights and choke.

Why you'd reach for it: consistency. The whole point of a preset block is that it's the same block every time - same phrasing, same emphasis - which means the style variable is held constant while you vary everything else. That's genuinely useful when you're batch-testing prompts, or when you want a stable baseline across a LoRA that's supposed to shift only one thing.

The honest caveats: presets are long, and every token you append is a token the model has to split attention across, so you can bury a specific instruction. Don't stack three bonus blocks. And a preset is a starting point, not a magic wand - "PHOTOGRAPHY_PHOTOREALISM" won't make a sketch model photoreal, it'll just fight the model's prior. Pick one block, treat the escape toggle as on-by-default for any text encoder that parses parentheses, and wire the output into your prompt concatenation.

Install is pack-standard: Manager search "ComfyUI-UtilsCollection" or git clone into custom_nodes. No models, no extra deps. This is the node the pack's preset family is built around - the video, camera-shot, and edit variants all follow the same dropdown-a-string template.