🎨 FoW - Style Category Light

You want the image to look like something - cyberpunk, oil painting, sumi-e, art deco - but you can never remember the three or four words that actually make a model lean that way. FoWLStyleAgent is the node that hands you a menu instead. It's the "Style Category Light" node from the FoW suite, a curated list of art styles where every style ships with its own ready-made descriptive tokens. You click the ones you want, it does the CLIP encoding, and out the other end comes a conditioning you can drop straight into a KSampler.

It's aimed squarely at beginners, and it shows - in the good way and the slightly awkward way.

How it works

Add the node and you'll see a clip input and a button that says "Open Style Catalogue". Clicking it opens a modal with a search bar and a long list of styles: Minimalism, Fluid Dynamics, Sumi-e, Art Deco, Cyberpunk, Anime, Photorealism, Space Opera, and so on. Expand a style and you get its individual tokens - each a short phrase with a tooltip explaining what it does ("Uses neon lighting", "Features bold lines", "Aims for realistic depiction"). Tick the ones you want.

Under the hood that's all the node ever does: it joins your selected tokens into a comma-separated string, feeds it through ComfyUI's standard clip.tokenize() + clip.encode_from_tokens_scheduled(), and emits a conditioning. No API, no model downloads, no weights. The catalogue itself is embedded as a hidden JSON widget, and the tokens are plain prompt text - which means your checkpoint's own CLIP encoder does all the work. If the model doesn't understand "neon", no node is going to fix that for you.

One honest caveat: the "tokens" are written to describe the style, but the image still depends on your checkpoint understanding them. On an anime SDXL finetune, "anime, expressive, vibrant, dynamic" barely registers because it's already the default. Where this node earns its keep is the styles you'd otherwise butcher - "bromoil print", "sumi-e", "tech noir" - where the curated phrasing genuinely beats whatever you'd type.

The inputs and outputs that matter

• clip - wire this from your checkpoint loader. It's what encodes the tokens.
• "Open Style Catalogue" button - this is the node's real interface. The user_input text field and the default_catalogue JSON are hidden widgets the button manages for you.

Outputs:

• Style Conditioning (CONDITIONING) - feed this into the KSampler's positive input (or a Conditioning (Combine) node if you're stacking categories).
• Style Text (STRING) - the literal comma-joined text it built. Useful if you want to see what got selected, or wire it into a regular CLIP Text Encode instead of using the conditioning output directly.

Installing it

The FoW suite has no dependencies beyond ComfyUI itself - no requirements.txt, no models to hunt down, nothing extra to download. Either grab it from ComfyUI Manager (search "Force of Will Suite Light"), or:

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

Then restart ComfyUI. That's genuinely it.

Common issues

• The text box is "missing". You'll right-click looking for the prompt field and find nothing - the user_input widget is hidden on purpose. The catalogue button is the input.
• Nodes not showing up after install. The usual ComfyUI cure: make sure the folder landed in custom_nodes/, not inside another folder, and restart fully.
• The catalogue feels thin. It is - the Light tier ships a starter list, and the README ties a bigger catalogue to the author's paid tiers. If you outgrow it, you can still type your own style tokens... you just can't get at the field without the modal.

The author is a self-described first-time coder who built this as his first project in two months, and it reads that way - quirky but earnest, MIT-licensed, and free of the dependency bloat that plagues bigger packs. For someone who wants "here's a picker, click the style, go", it does exactly that.