DiffusionBench: On Holistic Evaluation of Diffusion Transformers

Hugging Face Daily Papers Papers

Summary

Researchers introduce NanoGen, a unified framework for training and evaluating diffusion transformers, and propose DiffusionBench, a holistic benchmark combining ImageNet class-conditional and text-to-image generation to better assess progress in generative modeling.

Diffusion transformer (DiT) research on image generation has converged to a single evaluation setup: class-conditional generation on ImageNet. While methods improve the FID and related metrics, it is increasingly unclear whether they reflect real progress in generative modeling. The natural alternative, i.e., text-to-image (T2I) generation, is perceived as too costly or inconvenient to train and evaluate and is often skipped. We argue that this perception no longer holds. We introduce NanoGen, a unified DiT training and evaluation framework. NanoGen matches state-of-the-art DiT baselines on ImageNet and, with 12 lines of configuration change, also trains competitive text-to-image models. It currently supports RAE, VAE, pixel-space, and MeanFlow diffusion methods under both ImageNet and T2I setups. Under NanoGen, training T2I requires comparable compute to ImageNet. After training 21 latent diffusion models with NanoGen, we observe that method ranking shows no strong correlation between ImageNet and T2I generation: Pearson correlation is between -0.377 and -0.580 across three metrics. This suggests that a method which improves class-conditional ImageNet FID may show no corresponding improvement on T2I, clearly indicating the necessity of evaluating DiTs on both tasks. To this end, we summarize ImageNet and text-to-image results, which yields DiffusionBench, a holistic benchmark for DiT research. We recommend reporting DiffusionBench in place of ImageNet alone: methods that improve DiffusionBench are more likely to reflect broader progress.
Original Article
View Cached Full Text

Cached at: 06/24/26, 05:47 AM

Paper page - DiffusionBench: On Holistic Evaluation of Diffusion Transformers

Source: https://huggingface.co/papers/2606.24888

Abstract

Researchers introduce NanoGen, a unified framework for training and evaluating diffusion transformers that demonstrates the need for comprehensive benchmarking beyond ImageNet class-conditional generation to assess true progress in generative modeling.

Diffusion transformer(DiT) research onimage generationhas converged to a single evaluation setup: class-conditional generation onImageNet. While methods improve theFIDand related metrics, it is increasingly unclear whether they reflect real progress in generative modeling. The natural alternative, i.e., text-to-image (T2I) generation, is perceived as too costly or inconvenient to train and evaluate and is often skipped. We argue that this perception no longer holds. We introduce NanoGen, a unifiedDiTtraining and evaluation framework. NanoGen matches state-of-the-artDiTbaselines onImageNetand, with 12 lines of configuration change, also trains competitive text-to-image models. It currently supports RAE, VAE, pixel-space, and MeanFlowdiffusion methodsunder bothImageNetand T2I setups. Under NanoGen, training T2I requires comparable compute toImageNet. After training 21latent diffusion modelswith NanoGen, we observe that method ranking shows no strong correlation betweenImageNetand T2I generation: Pearson correlation is between -0.377 and -0.580 across three metrics. This suggests that a method which improves class-conditionalImageNetFIDmay show no corresponding improvement on T2I, clearly indicating the necessity of evaluatingDiTs on both tasks. To this end, we summarizeImageNetand text-to-image results, which yieldsDiffusionBench, a holistic benchmark forDiTresearch. We recommend reportingDiffusionBenchin place ofImageNetalone: methods that improveDiffusionBenchare more likely to reflect broader progress.

View arXiv pageView PDFProject pageGitHub36Add to collection

Get this paper in your agent:

hf papers read 2606\.24888

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.24888 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.24888 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.24888 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Diffusion Language Models: An Experimental Analysis

arXiv cs.AI

A systematic experimental analysis evaluating eight state-of-the-art Diffusion Language Models across multiple benchmarks, analyzing trade-offs between generation quality and computational efficiency.