Composition

Composition is the section of the Prompt Builder pack that handles how the image is framed rather than who is in it - camera, angle, shot size, quality tags, and even how many people are on screen. It's the node you reach for when every image you generate has the same flat, eye-level, full-body look and you want that to stop.

The pack's framing here is smart. Instead of a freeform "cinematic shot" text box, you get a 35mm lens and a low angle and an art-deco style modifier as separate dropdowns. On the anime-SDXL models this pack targets, those are Danbooru-flavored tags the model actually knows, so picking "low angle" actually changes the picture in a way typing the same words casually often doesn't.

The inputs that matter

14 dropdowns, but three clusters cover most of what you'll ever want:

• Quality tags - quality_1 and quality_2 offer score_9, score_8_up, score_7_up, masterpiece, best quality. The score_x family is Pony's convention: on Pony-trained checkpoints those tags genuinely steer output quality, which is why the pack ships with them. On an Illustrious or NoobAI checkpoint, masterpiece and best quality are the ones that carry weight. Either way, set one.
• Camera and framing - camera (canon eos r, 35mm/50mm/85mm lens), angle (low angle, from above, dutch angle…), and shot_emphasis (close-up, medium shot, full body shot). Naming a physical camera is one of the highest-leverage phrases in modern prompting, so this dropdown punches above its weight.
• Who's in frame - actors_1 (1girl, 2girls, 1guy, mixed combos) and group/solo (solo, duo, group, crowd). Yes, that second field really does have a slash in its name. They cover overlapping ground, so pick one and leave the other on None.

The rest - style_modifers (that spelling is real, it's a typo in presets.json that shipped, and since you pick from a dropdown it's purely cosmetic), color_palette, pov (male/female), three render_detail slots, and art_finish (cinematic lighting, ray tracing) - is texture. The render_detail trio is another case of duplicate slots: set one, leave two on None.

The output is composition - a string wired into the composition input on the pack's Combine Prompt node, where it lands second in the join order, right after the style preset and before the character. That's a sensible spot: framing gets strong attention but doesn't crowd out who's in the shot.

Install

No dependencies, no model files - this pack is plain Python and a JSON file. Clone it into custom_nodes or install via ComfyUI Manager (search "Prompt Builder"), then restart ComfyUI.

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

Common issues

The quality-tag dropdown is the main thing people trip over: the Pony score_9 convention doesn't transfer everywhere, so if you're on a non-Pony checkpoint, expecting score_9 to do magic is a recipe for disappointment - switch to masterpiece. And don't set all three render_detail slots; "highly detailed, ultra detailed, 8k resolution" stacked together is the kind of prompt bloat the pack's own README warns against. Pick your camera, your angle, one quality tag, and let the model do its job.