This LoRA was developed after several days of experimenting to achieve the highest possible quality from HiDream, particularly within ComfyUI workflows. My goal has been to refine a workflow that consistently delivers strong results, but achieving reliable high-res outputs with HiDream remains challenging.
The LoRA is specifically trained on close-ups and skin details, aiming to produce more natural-looking skin textures in generated images. It is best suited for scenarios where realistic skin and facial details are important.
Usage Recommendations:
For most images, I recommend using a LoRA weight of 0.5 or lower. At higher weights, the LoRA can restrict the base model’s creativity and variety.
Higher weights (above 0.5) should be reserved for close-up portraits where enhanced skin detail is desired.