<h1 id="style-transfer-(controlnet+ipa-v2)-bit2kpa7i">Style Transfer (ControlNet+IPA v2)</h1><h3 id="from-v1.3-onward-workflow-functions-for-both-sd1.5-and-sdxl-fpahc4oi4">From v1.3 onward workflow functions for both SD1.5 and SDXL</h3>How it works:This Worklfow will use 2 images, the one tied to the ControlNet is the Original Image that will be stylized. The Image loaded into IPAdapter will be used to stylize the original as well as possible. This will allow you to change the look of images quickly and efficiently. All you need to do is find the style image you want to use. A lot of the nodes in this workflow are node-groups, combining the effects of different nodes together to make an as small and efficient workflow as possible without a lot of "cable management".Important information about the Unified Image Loader (ControlNet): It has 2 distict Image Outputs. If you are working with a Monochrome Lineart Image you want to use the Lineart Output. If you use an Image like a Digital Art Image or Photograph that has less distinct lining for the Lineart ControlNet to detect you want to link up the Non-Lineart Image Output.This is because the Images need to be prepared for ControleNet. If you are already using a Lineart Picture it'll be inverted only. If it's a different sort of image the Node will automatically convert it to a Lineart picture in the background and use that for the ControlNet Adapter.You can go crazy and try it out both ways, but overal the output quality is better when you stick to the correct Image Outputs.What is needed:<ul><li>As always, I suggest you Download ComfyUI Manager by Lt.Dr.Data for easy instalment of any additional nodes you currently lack: <a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Manager">https://github.com/ltdrdata/ComfyUI-Manager</a></li><li>For general flexibility I suggest the Proteus Model (Linked Below)</li><li>The Checkpoint loader also has a LoRA slot for a Detail Adder. (I'll link the one I use Below)</li><li>You will need the Lineart Controlnet for SDXL (Provided in the Files *.zip)</li><li>For Flexibility I also Suggest 2 upscaler Model 1 for Artistic Purposes and 1 for Photographic Outcomes (Both are also in the Files Download)</li></ul>Where to put the Files from the *.zip:<ul><li>t2i-adapter-lineart-sdxl-1.0.safetensors goes into the ComfyUI\models\controlnet folder.</li><li>The 2 Upscale Models go into the ComfyUI\models\upscale_models folder</li></ul>VRAM Requirements:<ul><li>The last node in the Image Generation Group is a hiresfixStyle node. If your system is not capable of running this reliably you can always delete it and use Ultimate SD Upscale instead as it's way lighter on the VRAM.</li><li>Normally any GPU with at least 8Gb VRAM should be able to run this workflow without issues.</li></ul>As always, I provided more Notes inside the Workflow, but all test images uploaded here are all made with the exact same settings. So, it's quite forgiving in that department. While there are never single settings for this sort of stuff it seems that the LineArt ControlNet and IPAdapter together make for a very user-friendly workflow.PS: While everything is designed for SDXL if you install the SD1.5 ControlNet files for Lineart there is no reason it wouldn't work with that version of ControlNet as well. PPS: The Lineart pictures used in the previews where all made with my UNET - Stable Cascade Workflow (Linked Below)

Style Transfer (ControlNet+IPA v2)

The Mad Scientist Update<ul><li>IPAdapter's latest update added the Mad Scientist Node to tap into Crossattention this function has now been added to the Workflow.<ul><li>More information can be found in Info Note in the Workflow.</li><li>Also it might be interesting to check out Matt3o's Video on the new Node: <a target="_blank" rel="ugc" href="https://www.youtube.com/watch?v=ewKM7uCRPUg">https://www.youtube.com/watch?v=ewKM7uCRPUg</a></li></ul></li></ul>

Files with SHA256: 506f432bc93093479468747dc1de16eceafa911ef5b79437c5333121a10df471