T-DIG lets users steer image diversity with natural-language prompts
Original title: T-DIG: Text-Conditioned Diversity for Image Generation
It keeps the image generator frozen and adds user-directed diversity to open and closed models without weight access.
T-DIG is a framework that lets users request image variations along specific axes, such as background or style, using natural language. An LLM policy, trained with GRPO using a diversity reward and alignment penalties, generates axis-conditioned prompts for a frozen generator or editor, so the approach works with open-weight and proprietary models. The authors also release a dataset for text-conditioned prompt diversity and report that T-DIG exceeds baselines while preserving base-prompt alignment.
Unconfirmed: Superiority over baselines is the authors' claim based on their metrics and human judgements.
- Uses an LLM policy trained with GRPO to produce axis-conditioned prompts or edit instructions.
- The generator or editor stays frozen, so the method applies to open-weight and proprietary models.
- Authors release the first dataset for text-conditioned prompt diversity in image generation.
From the original article
1 min read at huggingface.co
Text-to-image (T2I) platforms typically return several images per prompt so users can explore different visual interpretations and choose one that matches their intent. However, T2I models are usually optimized for single-image quality which comes at a cost of multi-image diversity, limiting this exploration.
By Simran Khanuja, Rohit Gandikota, Gaurav Parmar · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.