This was an experiment that just seems to work, I really don't know how or why. It seems that interpolation of latents with Flux yields more fine details in images. To vary an image more substantially you can try adding a node to set another seed for the 2nd pass, this allows you to change the image details while retaining quality and most of the composition. I haven't explored other types of styles with this workflow besides photos.
I CANNOT PROVIDE SUPPORT FOR THIS, I'M JUST SHARING!
This workflow uses araminta_k_flux_koda.safetensors which can be found at CivitAI.
https://civitai.com/models/653093/Koda%20Diffusion%20(Flux) -- Amazing lora!
The Kodachrome LUT file can be downloaded from here: https://www.freepresets.com/product/free-lut-kodachrome-lookup-table/
LUT goes into ComfyUI\custom_nodes\ComfyUI_essentials\luts, if you are using cubiq's comfy_essential nodes like in this workflow.
The Flux.1 checkpoint used in this workflow is the dev version. If you're missing any custom nodes or get errors/red nodes:
- Click manager
- Click on "Install Missing Custom Nodes"
- When the installation finishes, click "Restart" as instructed
- Reload the workflow
I'm using an RTX 4090 with 24GB of RAM. Each image takes approximately 98 seconds.
If you modify and share this workflow, all I ask is that you credit me. https://renderartist.com https://www.instagram.com/renderartist/ https://www.x.com/renderartist/