Diffusion Model Inversion With Iterative Refinement

via “refiner model integration for iterative quality improvement”

text-to-image model by undefined. 20,41,667 downloads.

Unique: Implements two-stage generation with separate refiner model that continues from base model latents, enabling optional quality improvement without increasing base model size; supports flexible composition of base and refiner for quality/latency tradeoff

vs others: More modular than single-stage models (refiner is optional); enables quality improvement without retraining base model; comparable to other two-stage approaches but with better integration and documentation

* ⭐ 11/2022: [Visual Prompt Tuning](https://link.springer.com/chapter/10.1007/978-3-031-19827-4_41)

Unique: Combines DDIM-based fast sampling with learnable text embeddings during inversion, allowing the inversion process itself to discover semantic representations that align with natural language. This is architecturally distinct from prior inversion methods that treat text as fixed or use only pixel-space reconstruction losses.

vs others: Faster and more semantically meaningful than naive pixel-space optimization because it leverages the diffusion model's learned semantic structure and text alignment, producing inversions that are more amenable to text-guided editing.