Today's Theme: Image Creator [Image to Image/image1/GPT-image2.0 x LLM x Upscaler]

We iterate the LLM to build scenes aligned with the theme and automatically generate and output prompts optimized for GPT-image2.0. We express compositions that are hard to achieve with standard image generation by leveraging image-to-image characteristics. To reduce cost, we upscale low-quality 1k output.
Note: Because GPT-Image2.0 has a strong tendency to output the same image as the original, we’ve added a prompt to suppress this.