🚀 Apply Style Model (Enhanced)

The Simple variant of this node is great until you realize you want a character's look and a painting's texture and a lighting mood in one image. Chaining three style applications means three sets of CLIP Vision + style-model wiring. This node is that chain collapsed: up to three reference images, each with its own enable switch and strength, all folded into a single CONDITIONING output.

It's the "enhanced" version the pack ships alongside ApplyStyleModelSimple, and mechanically it's the same Redux-family trick - CLIP Vision encodes each image, the style model turns it into conditioning, and everything gets concatenated onto your text conditioning. What's new is the multi-input plumbing and the per-image switches, which let you toggle references on and off without ripping wires out of the graph.

How it works

For each of the three optional image slots, a switch_1/2/3 boolean gates whether that image is processed at all. If a switch is off (the default for all three), the slot is skipped. Each enabled image gets CLIP-encoded, style-conditioned, and multiplied by its own strength_1/2/3, then all enabled style sequences are concatenated into one and appended to the incoming conditioning.

Two details worth knowing before you trust it blindly:

• If no switch is on, the node is a pass-through. You get your original conditioning back untouched. Handy for toggling style on/off in an A/B setup; also a silent failure if you thought image_1 alone would work - the switch defaults to off.
• The attn_bias strength_type is a trap here. The Simple node implements attention-bias strength properly; this one only applies the per-image strengths when strength_type is multiply. Select attn_bias and your strengths are silently ignored - the styles all join at full strength. Use multiply (the practical choice for multiple references anyway).

Inputs and outputs

Required: conditioning, clip_vision, style_model, plus crop (center/none) and strength_type (multiply/attn_bias). Optional: switch_1/2/3 (default off), image_1/2/3, strength_1/2/3 (0–10, default 1.0).

Output is one CONDITIONING, ready for the sampler.

Install

From ComfyUI Manager (search "YarvixPA") or

cd ComfyUI/custom_nodes
git clone https://github.com/YarvixPA/ComfyUI-YarvixPA

restart, and it's under ComfyUI-YarvixPA/Flux/Redux. As with the Simple node you supply the Redux-style model and CLIP Vision yourself.

When to actually use it

Multiple style references are a real workflow for people blending a subject reference with an environment style - character consistency from one image, palette from another. But be honest about the failure mode: stacking three styles at full strength is a recipe for mush. Leave two switches off, dial one reference in low, and only bring the others in if the single-style result is missing something. The switches make that experimentation cheap, which is the node's best quality.