Text-to-Image Generation without Training
One-Prompt-One-Story (1Prompt1Story), an innovative tool presented at ICLR 2025, distinguishes itself from traditional image generation methods that require large amounts of training data. The tool employs a breakthrough no-training technology paradigm that enables high-quality image generation based directly on existing text embedding models.
The core technological innovation lies:
- Singular value reweighting technology: achieve better semantic understanding by intelligently reweighting the singular values of the text embedding matrix
- Identity Preserving Cross Attention Mechanism: Ensuring Stability of Role Identities During Multiple Generation Rounds
The biggest advantage of this approach is that it avoids the cost of model fine-tuning while maintaining image consistency and quality performance that is comparable to or even better than the trained approach.
This answer comes from the articleOne-Prompt-One-Story: Text Prompts Generate Character Identity Consistent ImagesThe
May not be reproduced without permission:AI productivity tools " One-Prompt-One-Story is a text-to-image generation tool that requires no training
