Project_Sticker_Design(LLM)
收藏官方服务：
资源简介：
Built a closed-loop auto animal sticker pipeline based on Stable Diffusion 1.5. Data & Fine-tune: Cleaned 3-style sticker datasets; train individual UNet LoRAs with DreamBooth for fixed visual styles. Generation & Feedback: Batch text/img2img generation. GPT-4oMini scores stickers on 5 dimensions; low-score samples auto-regenerate via parameter tuning. Evaluation: Tested 111 stickers (avg score 8.03); objective image metrics rank Digital Art Cartoon as top style. Post-process: GPT-4o generates short captions, renders sticker border layers, exports ready-to-use PNG. Outlook: Optimize layout centering via spatial prompt constraints and custom sticker evaluation metrics.
提供机构：
Zenodo创建时间：
2026-07-25
