3,000,000 text-to-image prompts curated for training and evaluating few-step generation models. The collection supports the pi-Flow framework developed by Stanford and Adobe Research for imitation distillation.
Use Cases
- Fine-tune diffusion models using the 3 million prompt entries to optimize for few-step inference
- Benchmark prompt-following capabilities of T2I models using the diverse text descriptions
- Conduct imitation distillation experiments by pairing these prompts with teacher model outputs
Strengths
- 3,000,000 text prompts for text-to-image (T2I) generation tasks
- Curated for the pi-Flow paper focusing on policy-based few-step generation
- Supports integration with pi-Qwen and pi-FLUX model architectures