Text-Guided Image Editing Based on Post Score for Gaining Attention on Social Media

Area of Science:

• Computer Vision
• Natural Language Processing
• Artificial Intelligence

Background:

• Text-guided image editing enables image modification based on textual descriptions.
• Current methods produce varied results based on prompt phrasing, requiring user selection.
• Social media sharing necessitates edited images that capture audience attention.

Purpose of the Study:

• To propose a novel text-guided image editing method optimized for social media engagement.
• To develop a system that automatically selects the most attention-grabbing edited image for social media platforms.

Main Methods:

• Utilized pre-trained text-guided image editing models to generate multiple edited images from various text prompts.
• Employed a large language model to create diverse text prompts for image editing.
• Developed and integrated a novel model to predict social media engagement scores for edited images.

Main Results:

• The proposed method generates edited images that accurately reflect the input text prompts.
• The system successfully identifies and selects images predicted to achieve higher audience engagement on social media.
• Experimental results on real Instagram data show a positive impression and improved attention compared to existing methods.

Conclusions:

• The novel approach enhances text-guided image editing for social media by prioritizing audience engagement.
• This method offers a practical solution for users aiming to maximize the impact of edited images online.
• The integration of engagement prediction advances the application of AI in content creation and social media strategy.