Tansan (Anime Portrait) LoRA for ZiT

Independent AI artist ThePoetPyronius has launched a new aesthetic tool for the image generation community: the 'Tansan (Anime Portrait)' LoRA model, specifically fine-tuned for the ZiT (Zero-shot Image-to-Text) model. This release provides a specialized style modifier that excels at generating highly stylized anime portraits. The model was trained for 10 epochs over 2,500 steps using CivitAI's LoRA trainer on a 1024p dataset, with a learning rate of 0.0005 and a cosine scheduler at rank 32. Its defining characteristic is a pronounced depth-scaling effect that artistically blurs both foreground and background elements, creating a visually striking, portrait-focused composition.

The creator notes this ZiT version is more heavily weighted towards portrait composition than a previous Qwen model variant, making it ideal for single-character shots but less effective for dynamic poses or scenes with multiple characters. The recommended strength for optimal results is between 0.6 and 0.8. A known trade-off of the fluid art style is occasional anatomical anomalies, particularly with hands at higher strengths. ThePoetPyronius suggests users experiencing this issue lower the model strength and employ classic positive prompts like 'best hands, five fingers' for correction. The model is now available for download, offering a new tool for artists seeking a specific, dreamlike anime portrait aesthetic.