{"slug": "test-time-scaling-of-diffusions-with-flow-maps", "title": "Test-time scaling of diffusions with flow maps", "summary": "Researchers propose Flow Map Trajectory Tilting (FMTT), an algorithm that uses a flow map's relationship to the velocity field to perform fast look-ahead during test-time scaling of diffusion models, and they claim it provably performs better ascent on user-specified rewards than standard test-time methods that inject the reward gradient into the diffusion dynamics. The method can sample exactly via importance weighting or run principled search for local maximizers of the reward-tilted distribution, and the authors demonstrate it against other look-ahead techniques and use it with vision language models for new forms of image editing.", "body_md": "A common recipe to improve diffusion models at test-time so that samples score highly against a user-specified reward is to introduce the gradient of the reward into the dynamics of the diffusion itself. This procedure is often ill posed, as user-specified rewards are usually only well defined on the data distribution at the end of generation. While common workarounds to this problem are to use an approximate denoiser to estimate what a sample would have been at the end of generation, we propose a simple solution to this problem by working directly with a flow map. By exploiting a relationship between the flow map and velocity field governing the instantaneous transport, we construct an algorithm, Flow Map Trajectory Tilting (FMTT), which leverages a fast yet precise flow map look-ahead and provably performs better ascent on the reward than standard test-time methods involving the gradient of the reward. The approach can be used to either perform exact sampling via importance weighting or principled search that identifies local maximizers of the reward-tilted distribution. We demonstrate the efficacy of our approach against other look-ahead techniques, and show how the flow map enables engagement with complicated reward functions that make possible new forms of image editing, e.g. by interfacing with vision language models.", "url": "https://wpnews.pro/news/test-time-scaling-of-diffusions-with-flow-maps", "canonical_source": "https://research.nvidia.com/publication/2026-12_test-time-scaling-diffusions-flow-maps", "published_at": "2026-11-01 03:00:40+00:00", "updated_at": "2026-11-01 03:22:30.471691+00:00", "lang": "en", "topics": ["generative-ai", "machine-learning", "computer-vision", "ai-research"], "entities": ["Flow Map Trajectory Tilting", "FMTT"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/test-time-scaling-of-diffusions-with-flow-maps", "markdown": "https://wpnews.pro/news/test-time-scaling-of-diffusions-with-flow-maps.md", "text": "https://wpnews.pro/news/test-time-scaling-of-diffusions-with-flow-maps.txt", "jsonld": "https://wpnews.pro/news/test-time-scaling-of-diffusions-with-flow-maps.jsonld"}}