Scholarlink

Image Synthesis is a method of generating new images, with significant applications in entertainment and media, such as creating game assets, concept art, and visual storytelling. This research integrates the CLIP and VQGAN models to generate visually coherent images based on textual prompts. By incorporating CLIP's ability to align visual and textual embeddings alongside VQGAN's high-resolution image synthesis capabilities, this approach creates realistic cartoon images that closely match the given textual descriptions. After the images are created, they are converted into videos for visual presentation. Experimental results demonstrate that the CLIP + VQGAN multimodal outperforms with other models such as Deep Dream, Pix2Pix, and DCGAN. The CLIP + VQGAN model achieves the highest inception score (5.0), the highest SSIM score (0.93), the lowest FID score (30), and improved precision and recall compared to existing models.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Creative Image Generation from Prompt Text Using CLIP and VQGAN Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • V. Umarani,
  • L. Prasanna,
  • S. Senthamizhan,
  • S. Dhanush

摘要

Image Synthesis is a method of generating new images, with significant applications in entertainment and media, such as creating game assets, concept art, and visual storytelling. This research integrates the CLIP and VQGAN models to generate visually coherent images based on textual prompts. By incorporating CLIP's ability to align visual and textual embeddings alongside VQGAN's high-resolution image synthesis capabilities, this approach creates realistic cartoon images that closely match the given textual descriptions. After the images are created, they are converted into videos for visual presentation. Experimental results demonstrate that the CLIP + VQGAN multimodal outperforms with other models such as Deep Dream, Pix2Pix, and DCGAN. The CLIP + VQGAN model achieves the highest inception score (5.0), the highest SSIM score (0.93), the lowest FID score (30), and improved precision and recall compared to existing models.