Test-time scaling of diffusions with flow maps
cs.LG, cs.AI
Submitted: 2025-11-27
Updated: 2026-11-07
Comments: Project page: https://flow-map-trajectory-tilting.github.io/
Code: https://github.com/black-forest-labs/flux
Project page: https://flow-map-trajectory-tilting.github.io/ABSTRACT
License: http://creativecommons.org/licenses/by-sa/4.0/
The gist: 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…
Terminology
Abstract
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 a 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 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.
Sources
- Stochastic Interpolants: A Unifying Framework for Flows and Diffusions
- Concrete Problems in AI Safety
- Qwen2.5-VL Technical Report
- Flow map matching with stochastic interpolants: A mathematical framework for consistency models
- How to build a consistency model: Learning flow maps via self-distillation
- Latent Beam Diffusion Models for Generating Visual Sequences
- Mean Flows for One-step Generative Modeling
- GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignment
- Consistency Trajectory Models: Learning Probability Flow ODE Trajectory of Diffusion
- Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps
- Flow Matching with Semidiscrete Couplings
- Test-Time Scaling of Diffusion Models via Noise Trajectory Search
- Align Your Flow: Scaling Continuous-Time Flow Map Distillation
- A General Framework for Inference-time Scaling and Steering of Diffusion Models
- Feynman-Kac Correctors in Diffusion: Annealing, Guidance, and Product of Experts
- Score-Based Generative Modeling through Stochastic Differential Equations
- Consistency Models
- Optimised Annealed Sequential Monte Carlo Samplers
- Skywork-VL Reward: An Effective Reward Model for Multimodal Understanding and Reasoning
- UniGenBench++: A Unified Semantic Evaluation Benchmark for Text-to-Image Generation
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis