rachanahegde
commented
8 months ago - User can design prompt to generate standalone images or images of a single consistent character
- For consistent character generation, they design the character first then generate images. I implemented few shot prompting to edit the prompts with GPT-4 and maintain the character's appearance while placing them in different settings.
- Freestyle image generation just uses the prompt editing available with Dall E 3 API
- Prompt-image pairs are displayed on page
- User can view edited prompts from GPT-4 and the final prompt used by Dall E 3 API (slightly edited even if I instruct it not to edit because this feature can't be turned off in the API)
- User can fill out form to control image generation via prompting
- User can save images by dragging and dropping to desktop or right clicking. (There's no explicit save option in the website.)