Regional Style Model Apply

RegionalStyleModelApply takes a reference image's visual style and injects it into your Flux conditioning - with a strength dial and the ability to layer it on top of an existing prompt's conditioning instead of replacing it. It's the pack's take on the stock StyleModelApply node, built for the pack's regional-prompting workflows. In plain terms: show it a photo of a moody neon alley and it will drag your generated scene in that direction, as hard as you tell it to.

What it is and isn't

Let's kill the confusion early: this node does not call an API, needs no key, and isn't Redux itself. It works with the standard ComfyUI STYLE_MODEL type - the same style-model mechanism the stock "Style Model Apply" node uses. You feed it a style model (loaded with a Style Model Loader, e.g. an InstantStyle-style model or a Redux-style model) plus a CLIP vision embedding of your reference image (CLIPVisionLoader → CLIPVisionEncode), and it produces conditioning the sampler can use.

The README's note is the honest one: you can combine the pack's regional prompting with Redux, but Redux is so strong it tends to dominate the generation. That's exactly the knob this node gives you - strength (default 1.0, range -1000 to 1000) - so you can bring the reference down to a whisper instead of letting it steamroll your prompt.

How to use it

Inputs:

• style_model (STYLE_MODEL) - from a Style Model Loader.
• clip_vision_output (CLIP_VISION_OUTPUT) - from CLIPVisionEncode on your reference image.
• strength (FLOAT, default 1.0) - how much the style conditioning counts. Below 1 is a light nod; around 1 is the default; above 1 starts dominating.
• conditioning (CONDITIONING, optional) - if given, the style tokens are concatenated onto your existing prompt's conditioning rather than replacing it. Leave it disconnected and you get a conditioning made purely of the style.

Output:

• CONDITIONING - wire this into your guider or into a CreateRegionalCond when used with the pack's regional workflow.

The signature move is feeding it into the regional pipeline: style-condition one region of the image and leave the rest on plain prompts, or run it at low strength on the whole canvas to nudge mood without losing prompt control.

Installation

Standard for the pack: ComfyUI Manager → search "ComfyUI-Fluxtapoz" → install → restart, or git clone https://github.com/logtd/ComfyUI-Fluxtapoz into custom_nodes. Only dependency is einshape; the style model and CLIP vision model are the ones you already load for normal style/reference work.

Common issues

• "The reference ate my prompt." That's strength too high (or Redux dominance, per the README). Dial strength down to 0.2–0.5 and the prompt reasserts itself.
• Wired it and nothing changes - check that the output conditioning actually reaches the sampler (through a guider), and that your style model's expected input matches how you encoded the reference. A style model that wants a particular CLIP vision config won't behave if given the wrong one.
• "It's just a stock node with extra steps?" Almost - the additions that matter are the strength multiplier and the conditioning concatenation. The stock node has neither.

For style-led edits, it's the cleanest way in this pack to say "make it look like this" with a volume control on how much this wins.