Preset Character

Preset Character is the Prompt Builder pack's shortcut node: 149 named characters from games and anime - Ada Wong from Resident Evil, Aerith Gainsborough from Final Fantasy VII, Ahsoka Tano, Albedo from Overlord - each one a single dropdown choice instead of a stack of tags you'd have to remember or look up. It's the fastest path in the whole pack from "I want that character" to "the model is trying to draw that character."

The catch is in that last phrase, and it's worth being honest about. Selecting "Ada Wong from Resident Evil" doesn't summon the actual Ada Wong - it produces a text string that describes her. Whether you get her depends entirely on whether your checkpoint has seen enough Resident Evil art to recognize the description. On a character-heavy anime checkpoint, that's often a solid hit. On a generic model, "Aloy from Horizon Zero Dawn" is just a loose descriptor and you'll get "a red-haired woman in a futuristic setting," which is not the same thing. For real consistency, pair this node with an actual character LoRA.

How it works

Two inputs. preset_character is the character dropdown itself (149 options). weight_style is the interesting one - it defaults to none but offers slight emphasis, medium emphasis, and strong emphasis, and here's what those actually do in the code: they wrap the whole character name in parentheses, once, twice, or three times.

none:              Ada Wong from Resident Evil
slight emphasis:   (Ada Wong from Resident Evil)
medium emphasis:   ((Ada Wong from Resident Evil))
strong emphasis:   (((Ada Wong from Resident Evil)))

That's standard attention weighting, and it comes with a caveat that matters more every year: the (...) syntax only does anything on CLIP-encoded models - the SDXL lineage, Pony, Illustrious, NoobAI. On 2026 LLM-encoded models like Flux, Z-Image, or Anima, the parentheses are discarded and you're just emphasizing nothing. If you're on those, leave weight_style on none and move on.

The output is preset_character - a string that feeds the preset_character input on the pack's Combine Prompt node. Note where it sits in Combine's join order: right after composition and before the full Character node's output, so the character name leads the detailed description. On CLIP models, where earlier tokens carry more attention, that ordering is doing real work for you.

Install

The standard pack install - no dependencies, no downloads, Python 3.10+.

cd ComfyUI/custom_nodes
git clone https://github.com/zinigo-creations/comfyui-prompt-builder

Or ComfyUI Manager → search "Prompt Builder" → install → restart.

Common issues

The two real failure modes are the wrong checkpoint and over-emphasis. On a checkpoint that doesn't know the character, no amount of (((...))) will fix it - the model literally has no concept to amplify. And cranking weight_style to strong on a character the model does know can start warping the whole image around her, crowding out every other tag in the prompt. Keep it at none or slight unless a specific character is fighting for attention. If you edit presets.json to add your own characters, restart ComfyUI to reload the dropdown.