# FIND: Fine-tuning Initial Noise Distribution with Policy Optimization for Diffusion Models

arXiv: https://arxiv.org/abs/2407.19453 · HTML: https://mlsanity.com/paper/2407.19453/

## Key facts

- Cited by 20 papers in the map (data through 14 Oct 2026).
- Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models.
- Builds on 2 of its 2 references that are in the map.
- First posted on arXiv 28 Jul 2024.

## Placed in

- Diffusion models / [Diffusion foundations and sampling](https://mlsanity.com/diffusion/diffusion-foundations-and-sampling/)

## Builds on (most cited first, 2 of 2)

- [LAION-5B: An open large-scale dataset for training next generation image-text models](https://mlsanity.com/paper/2210.08402/) (2022)
- [360^circ Reconstruction From a Single Image Using Space Carved Outpainting](https://mlsanity.com/paper/2309.10279/) (2023)

## Built on by (most cited first, 20 of 20)

- [Reward Fine-Tuning Two-Step Diffusion Models via Learning Differentiable Latent-Space Surrogate Reward](https://mlsanity.com/paper/2411.15247/) (2024)
- [The Silent Assistant: NoiseQuery as Implicit Guidance for Goal-Driven Image Generation](https://mlsanity.com/paper/2412.05101/) (2024)
- [Physics-Informed Distillation of Diffusion Models for PDE-Constrained Generation](https://mlsanity.com/paper/2505.22391/) (2025)
- [Noise Matters: Optimizing Matching Noise for Diffusion Classifiers](https://mlsanity.com/paper/2508.11330/) (2025)
- [Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models](https://mlsanity.com/paper/2510.08625/) (2025)
- [FeRA: Frequency-Energy Constrained Routing for Effective Diffusion Adaptation Fine-Tuning](https://mlsanity.com/paper/2511.17979/) (2025)
- [ConsistentRFT: Reducing Visual Hallucinations in Flow-based Reinforcement Fine-Tuning](https://mlsanity.com/paper/2602.03425/) (2026)
- [Early Timestep Zero-Shot Candidate Selection for Instruction-Guided Image Editing](https://mlsanity.com/paper/2504.13490/) (2025)
- [You've Got a Golden Ticket: Improving Generative Robot Policies With A Single Noise Vector](https://mlsanity.com/paper/2603.15757/) (2026)
- [Antithetic Noise in Diffusion Models](https://mlsanity.com/paper/2506.06185/) (2025)
- [FASTER: Value-Guided Sampling for Fast RL](https://mlsanity.com/paper/2604.19730/) (2026)
- [COT-FM: Cluster-wise Optimal Transport Flow Matching](https://mlsanity.com/paper/2603.13395/) (2026)
- [TAUE: Training-free Noise Transplant and Cultivation Diffusion Model](https://mlsanity.com/paper/2511.02580/) (2025)
- [NoiseRater: Meta-Learned Noise Valuation for Diffusion Model Training](https://mlsanity.com/paper/2605.08144/) (2026)
- [StressDream: Steering Video World Models for Robust Policy Evaluation and Improvement](https://mlsanity.com/paper/2606.00267/) (2026)
- [Retrieving and Refining Winning Noise Tickets for Diffusion-Based Motion Generation](https://mlsanity.com/paper/2607.06843/) (2026)
- [Spectral Evolution Search: Efficient Inference-Time Scaling for Reward-Aligned Image Generation](https://mlsanity.com/paper/2602.03208/) (2026)
- [Na"ive PAINE: Lightweight Text-to-Image Generation Improvement with Prompt Evaluation](https://mlsanity.com/paper/2603.12506/) (2026)
- [VANDERER: Map-Free Exploration using Future-Aware and Visual-Curiosity-Guided Diffusion Policy](https://mlsanity.com/paper/2606.14879/) (2026)
- [ExploreNet: Learning Where to Explore in Diffusion GRPO](https://mlsanity.com/paper/2609.38329/) (2026)

## Abstract (arXiv)

In recent years, large-scale pre-trained diffusion models have demonstrated their outstanding capabilities in image and video generation tasks. However, existing models tend to produce visual objects commonly found in the training dataset, which diverges from user input prompts. The underlying reason behind the inaccurate generated results lies in the model's difficulty in sampling from specific intervals of the initial noise distribution corresponding to the prompt. Moreover, it is challenging to directly optimize the initial distribution, given that the diffusion process involves multiple denoising steps. In this paper, we introduce a Fine-tuning Initial Noise Distribution (FIND) framework with policy optimization, which unleashes the powerful potential of pre-trained diffusion networks by directly optimizing the initial distribution to align the generated contents with user-input prompts. To this end, we first reformulate the diffusion denoising procedure as a one-step Markov decision process and employ policy optimization to directly optimize the initial distribution. In addition, a dynamic reward calibration module is proposed to ensure training stability during optimization. Furthermore, we introduce a ratio clipping algorithm to utilize historical data for network training and prevent the optimized distribution from deviating too far from the original policy to restrain excessive optimization magnitudes. Extensive experiments demonstrate the effectiveness of our method in both text-to-image and text-to-video tasks, surpassing SOTA methods in achieving consistency between prompts and the generated content. Our method achieves 10 times faster than the SOTA approach. Our homepage is available at urlhttps://github.com/vpx-ecnu/FIND-website.