Actually, I had this workflow since Flux Model, but decided not to release it because it was very slow and heavy on processing power. However, all of those limitations have been unlock with Z-image Turbo, It works good result and very fast.
V3 - Prehook & Triple LoRA
Building upon Version 2, it's slightly slower but can still display up to three people in an image.
V2 - Prehook & Multi LoRA
To add more variety to your visuals, Prehook allows you to mix and match LoRA styles of clothing or POV poses. Experiment with them yourself; I can't say which is the best.
It seems the sample images of the LoRA anime outfits have been blocked, so currently only LoRA style poses are available as examples.
V1 - Multi LoRA
Description
FAQ
Comments (4)
this works really well for my use case thank you. I would love to see a version for 4 characters (hint hint)
Thanks for the review, but unfortunately, my GPU doesn't have enough VRAM for more complex tasks. If you're interested, you can try building upon my workflow. The key part seems to be masking; perhaps combining it with ControlNet could improve individual positioning in the image.
@TtitanAI Thank you for the reply, I do appreciate it. I am fortunate enough to have a 5090, but I am quite new to comfyui. I would have loved to expand your great work but it took me a while to figure out how to use it never mind alter it. Maybe one day, thanks again.
This is an excellent work-flow that makes Zimage-Turbo a lot more versatile :)
Previously you could pretty much not have more than a single person (Or single face rather) in a zimage-turbo image, but this flow allows it.
It's all about finding a prompt and seed that places the 2 people within their own mask-zone, so they don't get blended (Though that can look quite interesting too. If a person is sitting in between the mask-zones he/she/it can get some pretty wild hair-styles :D So this flow can also be fun for experiments into punk-looks ;) )
It's a long-standing problem that it's so difficult to mix characters and keep them looking individual in the same photo, without resorting to trickery like inpainting or editing, so this workflow is very welcome.
Well done! :)