StyleModelApply_motorway_edition

Style models are a niche but lovely thing in the SDXL world: small extra models that inject a consistent look - a painter's palette, a film stock, a material texture - into your conditioning without retraining or a LoRA. StyleModelApply is the node that applies one, and it takes three inputs to do it: your conditioning, the style model, and a CLIP vision output that tells the style model what visual reference to draw on. This is the Motorway edition, so all three come out of the pipe by key.

Mechanically, the style model re-encodes your conditioning through a learned style transformation, using the CLIP vision output as the visual anchor. That's why strength_type offers multiply and attn_bias - two different ways the style gets fused into the conditioning, and which one works better depends on the style model and the checkpoint. If a style model seems to do nothing or nuke your image, switching between those two is one of the first things to try.

What to set

• INPUT_conditioning_key (default conditioning) - your positive conditioning from the CLIP text encode.
• INPUT_style_model_key (default style_model) - the STYLE_MODEL from a style model loader.
• INPUT_clip_vision_output_key (default clip_vision_output) - the CLIP_VISION_OUTPUT from a CLIP vision encoder, used as the style reference.
• strength (default 1, 0–10) - how hard to apply the style. 1 is standard; dial down for a hint of it.
• strength_type - multiply or attn_bias. Try multiply first; switch if the style feels off.
• OUTPUT_CONDITIONING_key (default CONDITIONING) - where the styled conditioning is written.

Output is the MOTORWAY with the styled CONDITIONING stored, ready for your KSampler.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Restart; it's under agilly1989 Nodes → Motorway-ed → conditioning → style_model. Manager has the pack. No pip dependencies.

Where people get burned

The CLIP vision output is easy to forget and the node will not remind you. Style model without a CLIP vision reference produces a weak or nonsensical result - the whole mechanism needs that visual anchor. Make sure a CLIP vision encoder ran and its output is in the pipe under the key you put in INPUT_clip_vision_output_key.

Then the standard Motorway trap: keys are case-sensitive and exact, and 'MotorwayClass' object has no attribute 'hash_' means a key isn't in the pipe - check the keys for conditioning, style_model, and clip_vision_output all exist upstream. And the pack is beta with motorway-ed nodes toggled across releases, so a node vanishing after an update is upstream churn, not a broken install.