Version 1.1: Added IP Adapter Noise node which sharpens up style details significantly. I didn't change it in the workflow, but setting a weight_style to 1.8 instead of 1 can apply the style more completely across the image, more than just the foreground. Like all of these settings, the benefit is subjective, so what may work best for one style input may work better without it on another input image.Version 1.0: This ComfyUI workflow is useful for rendering a Flux image and then applying a style over it with SDXL's IP Adapter and Controlnet ability.Adjusting the end percent on the Apply Controlnet node from 0.5 to 0.7 can increase how closely the image adheres to the composition of the original flux image. The controlnet model that's used can be downloaded here:<a target="_blank" rel="ugc" href="https://huggingface.co/xinsir/controlnet-tile-sdxl-1.0">https://huggingface.co/xinsir/controlnet-tile-sdxl-1.0</a>The IP adapter models and various other nodes that are non-standard can be downloaded inside ComfyUI's model and node manager.I'm using Flux Dev with a niji anime lora:<a target="_blank" rel="ugc" href="https://civarchive.com/models/871757?modelVersionId=1042979">https://civarchive.com/models/871757?modelVersionId=1042979</a>The SDXL model I'm using is Jibmix:<a target="_blank" rel="ugc" href="https://civarchive.com/models/194768?modelVersionId=672019">https://civarchive.com/models/194768?modelVersionId=672019</a>

Flux to SDXL Style with IP Adapter

Added IP Adapter Noise node which sharpens up style details significantly. I didn't change it in the workflow, but setting a weight_style to 1.8 instead of 1 can apply the style more completely across the image, more than just the foreground. Like all of these settings, the benefit is subjective, so what may work best for one style input may work better without it on another input image.

Files with SHA256: 2fc2e7905282f87827b2b50ec3819f6d50c88c192b8549305805b9c82d822b66