Efficient Image Synthesis with Sphere Latent Encoder
For researchers in few-step image generation, this work addresses computational inefficiency and objective conflict in Sphere Encoder, offering a more efficient and specialized framework.
The paper decouples Sphere Encoder into a fixed pretrained image encoder and a separate latent denoising model trained in spherical latent space, eliminating repeated pixel-space operations. On Animal-Faces, Oxford-Flowers, and ImageNet-1K, the method significantly outperforms Sphere Encoder in generation quality and inference speed, achieving competitive results against few-step and multi-step baselines.
Few-step image generation has seen rapid progress, with consistency and meanflow-based methods significantly reducing the number of sampling steps. Despite their low inference cost, these approaches often suffer from training instability and limited scalability. Sphere Encoder is a recent alternative that produces high-quality images in only a few steps; however, it requires repeated transitions between the pixel space and latent space during inference while jointly optimizing reconstruction and generation within a single architecture. This design leads to computational inefficiency and objective conflict between reconstruction and generation. To address these limitations, we decouple the framework into a fixed pretrained image encoder and a separate latent denoising model trained entirely in a spherical latent space. Our approach eliminates repeated pixel-space operations during training and inference, improving efficiency and allowing reconstruction and generation to specialize independently. On Animal-Faces, Oxford-Flowers and ImageNet-1K datasets, our method significantly outperforms Sphere Encoder in both generation quality and inference speed, while achieving competitive results against strong few-step and multi-step baselines.