AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Finite Difference Flow Optimization for RL Post-Training of Text-to-Image Models

David McAllister, Miika Aittala, Tero Karras, Janne Hellsten, Angjoo Kanazawa, Timo Aila, Samuli Laine

ECCV 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
switch Flow-GRPO to the EDM-style stochastic sampler (ablation)Flow-GRPO · Combined reward ablationreward convergenceImproved
replace Flow-GRPO policy gradients with finite difference flow optimization (ablation)Flow-GRPO with new sampler · Combined reward ablationreward convergenceImproved
start all trajectories in a group from the same initial noise (ablation)FDFO · Combined reward ablationstability and rewardImproved
use more unique prompts per epoch since only two rollouts per prompt are needed (ablation)FDFO · Combined reward ablationreward convergenceImproved
deterministic sampling for one trajectory of the pair instead of stochastic for both (ablation)FDFO · Combined reward ablationreward convergenceNo clear effect
use PPO clipping instead of SPO (ablation)FDFO · Combined reward ablationreward convergenceNo clear effect
add normalization of the difference vector (ablation)FDFO · Combined reward ablationtraining stabilityImproved
use true reward gradient via backprop through reward model instead of finite difference (ablation)FDFO · Combined reward ablationreward convergenceNo clear effect
backpropagate reward gradient through the sampling trajectory for per-step updates (ablation)FDFO · Combined reward ablationreward convergenceHurt
use FDFO instead of Flow-GRPO (baseline 40 steps)SD3.5-M LoRA · Combined reward wall-clockGPU hours to target reward: 19x fasterImproved
use FDFO instead of Flow-GRPO (fast 10 steps)SD3.5-M LoRA · Combined reward wall-clockGPU hours to target reward: 5x fasterImproved
use combined reward instead of PickScore aloneFDFO and Flow-GRPO · OneIG-Bench alignmentprompt alignmentImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy