-UPDATES- Added flexible model switching before the Face Detailers. You can now start with alternate base models like Illustrious or Pony to take advantage of their unique strengths and LoRA compatibility. Then, right before detailing, the workflow switches back to an SDXL model via a dedicated <code>Model Switch (JPS)</code> node and separate <code>Power LoRA Loader</code>. This ensures more accurate character restoration—especially useful for realistic SDXL-based character LoRAs, which tend to break on other bases.The model switch is hardwired to slot 2 for this SDXL return step—slots 3–5 are inactive for now. This design also helps reduce initial LoRA overload and “overcooking,” allowing you to delay full character rendering until the end of the pass. Still, it’s recommended to keep the character LoRA lightly active at the start (around 0.4–0.6) to establish base structure before refining.If you'd rather skip this handoff entirely, just disable “Enable SDXL Face Detailer” in the <code>Detailer Switch</code> node—everything will process on your original model and LoRA stack instead. Replaced the old Efficiency LoRA Stacker with the Power LoRA Loader (rgthree). The old node was a cluttered nightmare—dozens of static input fields whether you needed them or not. The new setup is clean, flexible, and way more manageable. You can toggle individual LoRAs, enable/disable them all at once, and dynamically add or remove layers as needed. Way less noise, way more control. Added a new Seed node to override the one built into the Sampler. The old version annoyed me—if you reloaded an image that had the seed set to -1 (random), it wouldn’t display the actual seed. This new node fixes that: it now correctly shows the seed for every image when reopened. General layout and quality-of-life updates. Reorganized the entire workflow into a smoother left-to-right layout, with all detailers now arranged side-by-side for faster access and cleaner review. Removed the preview window to declutter the interface. Note: At the moment, “Face Detailer 1” must be enabled for any of the other detailers (hands, feet, etc.) to function. This is because it currently initiates the shared detailer-pipe connection that feeds model and VAE data downstream. I’ll refactor this soon to make the pipe independent, but it’s low priority unless someone specifically needs the non-face detailers without the face pass. In practice, it rarely affects anything, since most workflows needing hands or feet also need a face. I've been a photographer/videographer for 20+ years and longed for something that felt a little closer to working in Lightroom or Capture One. It's not quite the same, but this workflow gives me a lot less anxiety. This doesn't have all the bells and whistles others have, but it's simple, clean, organized, and covers all the basics. Simple on/off switches for sections depending on the type of image you're creating based on simple categories, (portrait/landscape; do you need detailers or just upscale?) The sections are separated by detailers for Face 1st Pass, 2nd Pass, Hand, and Feet, and a 5th 'Upscale Only' section for images without people. I combined a lot of nodes because the chaos of hundreds of nodes drove me insane. The most important one to mention is probably that the Upscaler and Save nodes are combined. If you have a section activated, it will save it's upscaled image. It's set to save based on date-seed-what it detailed; subfolder - %date:MM-dd% filename - %date:MM-dd%_%seed%_face1 Hand and Feet detailers are hit or miss, would love any feedback on settings for those as I am still trying to find what works best. They do occasionally detail a hand into a foot, or a foot into a hand. It's hard getting a perfect bbox detection setting as the variables change image to image, so you may just need to tweak those settings as needed. There's also a "Preview" section on the right, this was just to have a section I could zoom my screen to nicely and leave it while it ran as I worked on my other computer. It's just 3 simple preview windows. Base output (step 1), after Face 2 (step 3), and after Feet (step 5).

Fenn SDXL Image Studio

Major Changes - Added flexible model switching before the SDXL face detailer. You can now start with alternate base models like Illustrious or Pony to take advantage of their unique strengths and LoRA compatibility. Then, right before detailing, the workflow switches back to an SDXL model via a dedicated <code>Model Switch (JPS)</code> node and separate <code>Power LoRA Loader</code>. This ensures more accurate character restoration—especially useful for realistic SDXL-based character LoRAs, which tend to break on other bases.The model switch is hardwired to slot 2 for this SDXL return step—slots 3–5 are inactive for now. This design also helps reduce initial LoRA overload and “overcooking,” allowing you to delay full character rendering until the end of the pass. Still, it’s recommended to keep the character LoRA lightly active at the start (around 0.4–0.6) to establish base structure before refining.If you'd rather skip this handoff entirely, just disable “Enable SDXL Face Detailer” in the <code>Detailer Switch</code> node—everything will process on your original model and LoRA stack instead. Replaced the old Efficiency LoRA Stacker with the Power LoRA Loader (rgthree). The old node was a cluttered nightmare—dozens of static input fields whether you needed them or not. The new setup is clean, flexible, and way more manageable. You can toggle individual LoRAs, enable/disable them all at once, and dynamically add or remove layers as needed. Way less noise, way more control. Added a new Seed node to override the one built into the Sampler. The old version annoyed me—if you reloaded an image that had the seed set to -1 (random), it wouldn’t display the actual seed. This new node fixes that: it now correctly shows the seed for every image when reopened. General layout and quality-of-life updates. Reorganized the entire workflow into a smoother left-to-right layout, with all detailers now arranged side-by-side for faster access and cleaner review. Removed the preview window to declutter the interface. Note: At the moment, “Face Detailer 1” must be enabled for any of the other detailers (hands, feet, etc.) to function. This is because it currently initiates the shared detailer-pipe connection that feeds model and VAE data downstream. I’ll refactor this soon to make the pipe independent, but it’s low priority unless someone specifically needs the non-face detailers without the face pass. In practice, it rarely affects anything, since most workflows needing hands or feet also need a face.

Files with SHA256: 2f7ed2de975101f6dff4fd155fa319a02a9aea0c3d0859e6003ad9aab3c9ee77