New post
Added 2025-09-05 12:45:06 +0000 UTCThese images have been generated using SDXL 1 at 1024×1024 resolution with 30 inference steps, running DPM++ 2M Karras for smoother detail rendering. CFG scale was set to 7.5, with a carefully tuned negative prompt to reduce artifacts and oversaturation. A secondary refiner pass at 20 steps was applied on the latent to improve facial consistency and micro-textures, followed by a 2× ESRGAN upscale and light denoising to maintain natural sharpness
