📸 Photo Style Helper

The outfit nodes handle what your subject wears. Photo Style Helper (class PhotoStyleHelperNode, in 👗 Outfit/Helpers) handles how the picture looks - and it's the node that keeps people who don't know a Rembrandt from a ring light from writing "nice lighting" and calling it a prompt. If you've ever stared at a prompt template that asks for lighting, lens, and depth of field and guessed, this node guesses for you, properly.

What it does

You pick a handful of photographic intentions from dropdowns - photo_genre (fashion, glamour, passport, street...), render_medium (photorealistic, or 2D anime / manga / cartoon / pixel art if you're off the photo path), lighting (softbox, golden hour, neon, Rembrandt...), lens (35/50/85/105mm), camera_framing (headshot to full-body), dof, composition, palette, and an outfit_vibe (modern, goth, pinup, techwear...). Then it emits four outputs you actually use:

• positive_additions - a comma-separated prompt fragment to append to your positive prompt (e.g. fashion photography, waist-up composition, softbox lighting, 85mm lens, shallow depth of field, studio quality).
• negative_additions - a matching negative fragment, and this is where it earns its keep. Passport mode adds tilted head, dramatic lighting, strong shadows, open mouth teeth; selfie/street adds motion blur; and any 2D medium automatically blocks photorealistic skin, skin pores, realistic skin texture, uncanny valley. That's the kind of detail people normally learn the hard way.
• preset_patch_json - a JSON patch meant for the outfit node, steering mood/time/background to match the genre. Fashion/editorial/runway pulls in dramatic studio vibes; portrait defaults to intimate mood at golden hour; passport goes flat and neutral.
• guidance_json - metadata plus sampler hints: quality_preset maps to steps/CFG/denoise/resolution (fast 20/4.5, balanced 30/5.5, high 40/6.5).

How to wire it

This node doesn't produce a final prompt on its own - it feeds the chain:

Photo Style Helper ──preset_patch_json──> Preset Patch Applier ──merged_json──> Male/Female Outfit Inputs From JSON ──> Outfit Node
positive_additions ──────────────────────────────────────────────────────────────────────> (append to your positive prompt)
negative_additions ──────────────────────────────────────────────────────────────────────> (append to your negative prompt)

Two honest caveats from reading the source. First, the patch sometimes sets values the outfit node's dropdowns don't officially contain (like time_of_day: studio) - the comment in the code calls this "harmless as a selector," and it mostly is, because the prompt builder just prints whatever string it's handed. Don't panic when you see an "unknown keys" warning in the console; check the output first. Second, the sampler hints are suggestions for you, not connections - the node doesn't touch your KSampler, it just tells you what the author would use. aspect_ratio also ends up folded into custom_attributes as a hint, not enforced anywhere, so set your actual canvas yourself.

Install and gotchas

Pack-standard install: ComfyUI Manager (search "ComfyUI Outfit Selection") or git clone https://github.com/manifestations/comfyui-outfit.git into custom_nodes/, restart. No models, no Ollama - all choices come from data/styles/photo_helper_options.json, which is a plain file you can edit to add vibes or lighting styles without touching Python. That's the node's quiet superpower: the whole thing is data-driven.

Where people get burned: expecting the negative additions to cover everything (they're heuristics, not a comprehensive negative), and forgetting that positive_additions is a fragment - if you append it to an already-sentenced Flux-style prompt you get a comma salad. On Flux, feed it as context, not as more tags.