Flux

First three hours with Z-Image Turbo as a fashion photographer

I shoot a lot of fashion photography and work with human subjects across different mediums, both traditional and digital. I’ve been around since the early Stable Diffusion days and have spent a *lot* of time deep in the weeds with Flux 1D, different checkpoints, LoRAs, and long i

Original ansehen ↗

The other posters were right. WAN2.1 text2img is no joke. Here are a few samples from my recent retraining of all my FLUX LoRa's on WAN (release soon, with one released already)! Plus an improved WAN txt2img workflow! (15 images)

Training on WAN took me just 35min vs. 1h 35min on FLUX and yet the results show much truer likeness and less overtraining than the equivalent on FLUX. My default config for FLUX worked very well with WAN. Of course it needed to be adjusted a bit since Musubi-Tuner doesnt have a

Original ansehen ↗

For some things, Z-Image is still king, with Klein often looking overdone

Klein is excellent, particularly for its editing capabilities, however.... I think Z-Image is still king for text-to-image generation, especially regarding realism and spicy content. Z-Image produces more cohesive pictures, it understands context better despite it follows promp

Original ansehen ↗