Normalizing Trajectory Models
We introduce Normalizing Trajectory Models (NTM), which models each reverse step as an expressive conditional normalizing flow with exact likelihood training.
Key points
- Diffusion-based models decompose sampling into many small Gaussian denoising steps, an assumption that breaks down when generation is compressed to a few coarse transitions.
- Its exact trajectory likelihood further enables self-distillation: a lightweight denoiser trained on the score function induced by the model itself produces high-quality samples in four steps.
- On text-to-image benchmarks, NTM matches or outperforms strong image generation baselines in just four sampling steps while uniquely retaining exact likelihood over the generative trajectory.
- In this work, we further advance the state of Normalizing Flow generative models by introducing iterative TARFlow (iTARFlow).
Sources (1)
- [1]Normalizing Trajectory ModelsApple Machine Learning Research · Oct 8, 12:00 AM
We introduce Normalizing Trajectory Models (NTM), which models each reverse step as an expressive conditional normalizing flow with exact likelihood training.
Diffusion-based models decompose sampling into many small Gaussian denoising steps, an assumption that breaks down when generation is compressed to a few coarse transitions.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 7, 2026Enabling Preference-driven Unlearning in Few-step Distilled Text-to-Image Diffusion Models
- Oct 7, 2026MUNITE: Unified Multimodal Latent Inference for Any-to-Any Multimodal Generation
- Oct 7, 2026Iris-3B: Going Beyond the Latent with Pixel-Space Diffusion Training, Conversion and Fine-Tuning
- Oct 7, 2026MIMESIS: Learning User Simulators as Training Environments for Interactive Agents
- Oct 6, 2026Consistent Distribution Matching for Data-Free Diffusion Distillation
- Oct 6, 2026Self-Retrospection Distillation: Turning Post-hoc Experiences into Prior Foresight