From Flux to SDXL, with folder batch.

The idea is to batch a folder of images. Find and complete the prompt, pass it to FLUX, then SDXL to add better texture/style, and fix the faces if necessary before upscaling. The original image is saved, too, for reference, with the prompt in a .txt. You are can add LORAs in each model, set Denoise (Flux, SDXL, Detailer), and save steps or not...

Sometimes FLUX is better, but it's rare :-)

I used Shuttle and DMD because they are fast, you can of course upgrade this to FLUX Dev and any SDXL model. This is perfect for Macs.

You can use this with any picture style, from an Instagram archive to a Simon Bisley anthology!

1) Folder Reader & Image to Prompt (Upper-Left Group)

• Loads images from a folder (random or incremental).

• Useful for batch processing.

• Converts an image into a rough text prompt (AI captioning).

• Provides a raw starting point.

2) Prompt Generation & Editing

• Sends the first prompt to Ollama (Gemma works well) (use Star Text Filter with DeepSeeker to remove unnecessary text).

• Add a Text C prompt if needed.

• All prompts are combined, saved as .txt, and sent to DMD2.

3) Primary Generation Group (Lower-Left)

• Uses Shuttle 3, a 4-step FLUX-based model, to generate an image.

4) Refinement Pass with DMD2 (Group in the Middle)

• Small upscale & detail enhancement with a second model pass (DMD2 is a 4-steps SDXL model).

• You can add a LORA loader here of course.

5) Detailer (Purple Group on the Right)

• Focuses on fixing faces.

6) Final Upscale & Save (Lower-Right)

• Upscales and adds slight film grain.

• Saves the final high-resolution image automatically.