🔍 FoW - Detail Category Light

The catalog shop for quality tags

FoW - Detail Category Light (the "Detail Agent") is the FoW suite's way of handling the detail/quality tags - highly detailed, 4k, sharp details, soft lighting, bokeh - without you memorizing them. Instead of typing, you click a button, get a catalog grouped by category (Quality, Resolution, Texture, Lighting, Detail Level...), tick the ones you want, and the node encodes them into conditioning. It's a prompt library wearing a node costume.

It's part of SirWillance's FoW_Suite_LIGHT, the free Light tier of the "Force of Will" suite - a self-taught dev's two-month first project, MIT-licensed, and explicitly aimed at beginners who'd rather click than type. The catalog is intentionally small in Light tier; the author's pitch is that higher paid tiers unlock bigger catalogs. Fine - the free one covers the essentials.

How it works

Mechanically the backend is the same for every "Agent" (positive category) node in the suite:

tokens = clip.tokenize(user_input)
cond = clip.encode_from_tokens_scheduled(tokens)
return (cond, user_input)

Your CLIP model encodes the text, the result becomes a CONDITIONING, and the same text comes out as a STRING. The interesting part is the frontend: the node hides the raw text widget and instead shows an "Open Detail Catalogue" button. Click it and a draggable modal lists the catalog - each token has a tooltip explaining exactly what it does ("Sharp image with very focused detail"). Tokens you tick get written back into the hidden user_input widget, which is what actually gets encoded.

The inputs that matter

• clip (CLIP) - the text encoder. Same CLIP your checkpoint uses; no separate model.
• user_input (STRING, multiline) - hidden on the node; the modal fills it with your selected tokens. You can also type straight into it if you prefer, but the point is the catalog.

Outputs are Detail Conditioning (CONDITIONING) and Detail Text (STRING). The conditioning feeds a positive Fusion node; the text is handy if you want to see or log exactly what got selected.

Installing it

One install gets the whole suite:

cd ComfyUI/custom_nodes
git clone https://github.com/SirWillance/FoW_Suite_LIGHT

Restart ComfyUI, and the nodes appear under "🧿 FoW - Light". ComfyUI Manager also has it - search "Force of Will Suite Light". No requirements.txt, no model downloads.

Where people get burned

The detail catalog is where generic quality tags live, and quality tags are exactly what the KB's prompt-engineering essay warns about: highly detailed, 4k is alive on SDXL-lineage checkpoints but largely inert on LLM-encoded models - and the whole FoW catalog does nothing to change that. Also don't tick everything. The modal's "Other" section (sharp focus, bokeh) is where people overdo it and end up with crunchy, over-textured output. Start with one or two, run, add more only if the image is actually under-detailed. One genuinely nice touch: the tooltips are honest, and you can wire the Detail Text output anywhere to see what the node is actually sending to the encoder.