🤖 AI Summary
Iterative generative models—such as diffusion models and flow matching—suffer from low inference efficiency due to hundreds of function evaluations per sample. To address this, we propose a warm-start mechanism that replaces the conventional random noise initialization with a context-conditioned prior distribution $mathcal{N}(mu, sigma)$, where the mean $mu$ and standard deviation $sigma$ are predicted in a single forward pass by a lightweight network. Conditional normalization enables model-agnostic, plug-and-play integration without modifying the original generator or sampler—ensuring compatibility across diverse iterative frameworks. In image inpainting, our method achieves performance on par with a 1000-step DDPM baseline using only one warm-start initialization plus ten sampling steps (11 total function evaluations), substantially accelerating strongly conditioned generation. Our core contribution is the first introduction of a context-aware, learnable initial prior into the general iterative generative paradigm.
📝 Abstract
Iterative generative models, like diffusion and flow-matching, create high-fidelity samples by progressively refining a noise vector into data. However, this process is notoriously slow, often requiring hundreds of function evaluations. We introduce the warm-start model, a simple, deterministic model that dramatically accelerates conditional generation by providing a better starting point. Instead of starting generation from an uninformed N(0, I) prior, our warm-start model predicts an informed prior N(mu, sigma), whose moments are conditioned on the input context. This "warm start" substantially reduces the distance the generative process must traverse, particularly when the conditioning information is strongly informative. On tasks like image inpainting, our method achieves results competitive with a 1000-step DDPM baseline using only 11 total function evaluations (1 for the warm start, 10 for generation). A simple conditional normalization trick makes our method compatible with any standard generative model and sampler without modification, allowing it to be combined with other efficient sampling techniques for further acceleration. Our implementation is available at https://github.com/jonas-scholz123/warm-start-model.