Wave Style Model Apply

Style models take a CLIP vision embedding of a reference image and warp your conditioning toward that image's look - think "this composition, in the style of that painting." The stock way to do it is a single application with a fixed strength. Wave Style Model Apply is the version that plays well with this pack's animation: it attaches the style model and the clip vision data to the wave_config, passes your conditioning through, and lets the samplers apply the style per-frame with strength riding the wave. The style saturates at the cycle's peak and lets the content show at the trough.

How it works

The node takes your conditioning, a loaded style_model, a clip_vision_output (from Wave CLIP Vision Encode), a strength type, and the ongoing wave_config. It clones the config, stashes the style model and clip vision data in it, and returns the conditioning unchanged for chaining. Just like its siblings in this pack, the "apply" is deferred - the samplers call into a style-model wrapper at sample time and modulate the strength per frame.

Two inputs to understand before the rest:

• strength_type - multiply (default) or attn_bias. Multiply scales the style conditioning by the strength factor; attn_bias injects the style as an attention bias, which is subtler and plays nicer with some models. If multiply looks too aggressive, try attn_bias before touching the wave.
• clip_vision_output - the reference look, encoded by Wave CLIP Vision Encode. If that node gave you a batch of references, the sequence rides along and cycles per cycle, so the style can walk from one reference to the next over the animation.

Outputs: conditioning (passthrough) and the enriched wave_config.

Where it sits

The README's style path: Wave CLIP Vision Encode → Wave Style Model Apply → Video/Image Iterative Sampler. This node's conditioning output continues to the sampler, and the enriched wave config keeps the style data and its modulation parameters on the same line. Style models load through ComfyUI's standard Style Model Loader; there's nothing pack-specific about the model itself.

Installing it

Part of kentskooking-nodes: ComfyUI Manager → search kentskooking-nodes → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Kentskooking/kentskooking-nodes

Restart after. No models to download - you bring the style model. Note this node does not need ComfyUI_IPAdapter_plus; that dependency is only for the IPAdapter nodes.

Common issues

• Style never appears - the wave config line got broken. The style data travels inside the enriched config; if the sampler receives an older, unstyled config, the style is simply never read. Keep one continuous line from controller through this node to the sampler.
• Style feels like it's fighting the prompt - switch strength_type to attn_bias, or lower the clip strength range in the base controller.
• It needs a wave_config input - unlike stock style-model nodes, there's no standalone mode. This is part of a wave chain by design; if you just want one static style application, the built-in Style Model Apply is the simpler tool.

One honest note: this node earns its keep in the iterative workflows, where per-frame modulation is the whole point. In a single static image, the stock apply node does the same job with fewer moving parts.