Ratatoskr (Qwen Image 2512 / Edit 2511) (Qwen Image (2512) V6.95)

Let my introduce the next Model Version Qwen Image (2512) V6.95 (Realistic) Like the model name suggest its focus is an more realistic output this is in fact an retrain trained over a week on an rtx6000pro blackwell gpu. the dataset of the model got extended massively. I added approx 1200 real images of real landscapes i visited and photographed myself over the years to get more realistic backgrounds. so what has changed: Training Resolution increased from 1024x1024 to 1328x1328 specific training on text added it now way better to difference from animals , humans and anthros sound trivial but its not ;-) Better contrast Realistic output Better in difference between sfw and nsfw the qwen grid is now almost gone added new kinks and stuff feel free to play with it ;-) drawings and comics still working but have an more detailed look for settings this model version didn't change' much from V2.99 onwards Sampler: Euler Scheduler: linear_quadratic or beta Steps between 25 and 35 CFG from 3-4 Workflow and examples are in the training folder. i use z-image for upscaling inside the workflow to avoid the qwen grid. so please give a like to support my work: