6arXiv cs.LG (Machine Learning)·
ROMS-IMLE: Single-step generative model achieves FID 2.56 on ImageNet 256 without iterative denoising
A new arXiv preprint introduces ROMS-IMLE, a minimalist single-step generative model that challenges the prevailing belief that iterative denoising (as in diffusion models) is necessary for high-quality image generation. Using Implicit Maximum Likelihood Estimation (IMLE) as the training objective and a convolutional network rather than a transformer, the model achieves FID 2.56 on ImageNet 256 with competitive precision and recall. The result is parameter-efficient and fast, suggesting that much of the complexity in modern generative pipelines may be unnecessary.
Related guides (1)
Related events (8)
Three-Body Scattering Modeling achieves state-of-the-art one-step image generation on ImageNet-256
A new arXiv preprint introduces Three-Body Scattering Modeling (TBSM), a generative modeling framework that uses energy distance to induce sample-level motion and provide regression supervision for one-step generators, without adversarial critics, prescribed noise paths, or autoregressive factorization. Each 'projectile' sample is attracted toward a real source and repelled from an independently generated source, with the expectation corresponding to the 2-Wasserstein gradient-flow velocity. On ImageNet-256, TBSM achieves FID=2.23 with PixelDiT-XL and FID=1.63 with latent DiT-XL at a single function evaluation (NFE=1), competitive with or surpassing multi-step diffusion methods. The work establishes a new design space connecting diffusion, drift-like, and GAN-like objectives for high-dimensional one-step generation.
Meta Research Improves Image Generation via Staged Planning and Self-Revision Fine-Tuning
Researchers from Meta and collaborating universities propose a fine-tuning method that teaches image generators to compose images through discrete plan-sketch-inspect-refine cycles rather than generating all at once. Starting from BAGEL-7B, they construct ~62,000 training examples using GPT-4o and FLUX.1 Kontext to supervise each stage, achieving 83% on GenEval versus 77% for the base model and a competing method (PARM) that required 11x more training data and ~8x more inference steps. The approach improves spatial relationship accuracy, object attribute fidelity, and real-world knowledge grounding in generated images.
Squeezing Capacity from MLLMs for Subject-driven Image Generation via Dual Layer Aggregation
This paper proposes conditioning diffusion models on Multimodal Large Language Models (MLLMs) that jointly encode text and reference images, augmented with VAE-based identity conditioning to address copy-paste artifacts and identity preservation failures in subject-driven image generation. A Dual Layer Aggregation (DLA) module aggregates multi-level MLLM features, and a multi-stage denoising strategy progressively balances semantic and fine-detail identity signals during inference. Experiments show improved human preference scores on subject-driven generation benchmarks compared to prior approaches that encode text and reference images separately.
Apple researchers propose Feature Auto-Encoder to speed diffusion training via compressed DINOv2 embeddings
Researchers at Apple introduced Feature Auto-Encoder (FAE), a latent diffusion image generator that compresses DINOv2 vision encoder embeddings before learning to denoise them, then expands them back for decoding. The approach achieves comparable image quality to state-of-the-art diffusion models while training roughly 7x faster on ImageNet class-conditional generation. The key insight is that shrinking semantically rich vision embeddings reduces compute during diffusion training without sacrificing the representational benefits of large pretrained encoders.
Improved Techniques for Training Consistency Models
OpenAI presents improved training techniques for consistency models, a class of generative models capable of producing high-quality samples in a single step without adversarial training. The work advances a nascent alternative to diffusion-based generation that trades multi-step sampling for single-step inference. The post originates from OpenAI's research blog, indicating continued investment in efficient generative modeling.
Glow: Better reversible generative models
OpenAI introduces Glow, a reversible generative model using invertible 1x1 convolutions that extends prior work on normalizing flows. The model generates realistic high-resolution images, supports efficient sampling, and learns disentangled features for attribute manipulation. Code and an online visualization tool are released alongside the paper.
FMLM+ introduces Posterior Refinement for fast non-autoregressive language generation
Researchers introduce FMLM+, a framework combining Flow Map Language Models with masking-style noise schedules to enable joint sequence generation with per-token global consistency scoring. The key contribution is Posterior Refinement, an inference-time self-correction strategy that matches discrete baseline performance with 32x fewer neural function evaluations (NFEs). The approach improves the speed-quality tradeoff over both Masked Diffusion Models and standard FLMMs across multiple benchmarks, addressing longstanding factorization error problems in non-autoregressive generation.
Image GPT: Transformer Models Applied to Pixel Sequences for Image Generation and Classification
OpenAI demonstrates that a large transformer model trained autoregressively on pixel sequences can generate coherent image completions and samples, analogous to text generation. The work establishes a correlation between generative sample quality and downstream image classification accuracy. The best generative model achieves features competitive with top convolutional networks in the unsupervised setting, suggesting shared representational principles across modalities.