The paper introduces techniques for efficient on-device diffusion-based image generation on consumer GPUs. It presents an embedding translator that reduces weight and latency by mapping a small text encoder into a larger encoder space, a reproducible sweep recipe for balancing speed, quality, and memory, and an interactive editor that achieves sub‑second time‑to‑first‑image on recent GPUs. These contributions aim to broaden the reach of diffusion pipelines to a wide range of client devices.
Large-scale visual generators are increasingly capable but costly to train, fine-tune, and deploy. We introduce Mage-Flow, a compact 4B-scale generative stack for efficient text-to-image generation and instruction-based image editing.
arXiv:2606. 13740v1 Announce Type: new Abstract: Diffusion large language models (dLLMs) accelerate generation by denoising multiple tokens in parallel, making them attractive for latency-sensitive mobile inference.
Flash-dLLM is a training‑free inference acceleration framework that improves the speed and memory efficiency of Diffusion Large Language Models (dLLMs). It tackles GPU memory I/O bottlenecks by introducing an I/O‑aware fused KV‑cache kernel and then employs a draft‑and‑verify decoding strategy that uses the dLLM itself as both drafter and verifier. Experiments on mathematical reasoning and code‑generation tasks show Flash‑dLLM outperforms existing acceleration methods, achieving up to 11.0× speedups over the Elastic‑Cache baseline.
arXiv:2607. 19064v1 Announce Type: cross Abstract: Large-scale visual generators are increasingly capable but costly to train, fine-tune, and deploy.
arXiv:2510. 03434v3 Announce Type: replace-cross Abstract: We present Paris, the first publicly released diffusion model pre-trained entirely through decentralized computation.
arXiv:2603. 29002v3 Announce Type: replace-cross Abstract: Modern large language models (LLMs) increasingly depends on efficient long-context processing and generation mechanisms, including sparse attention, retrieval-augmented generation (RAG), and compressed contextual memory, to support complex reasoning.
LeanStream introduces a speculate‑and‑refine streaming framework that enables efficient on‑device inference of large language models by progressively refining computation, loading, and cache‑retention priorities using partial GPU results. This approach allows fine‑grained overlap between GPU execution and storage I/O, avoiding the trade‑offs of existing systems that serialize execution or incur redundant weight fetches. Implemented on mobile and embedded platforms, LeanStream reduces memory usage by 4.8× to 7.5× compared to prior work while improving token generation throughput by 1.6× to 2.1×.
DC-Gen is a post‑training framework that accelerates text‑to‑image diffusion models by using a deeply compressed latent space. It first aligns the base model’s latent representations with a lightweight embedding alignment, then applies minimal LoRA fine‑tuning to preserve generation quality. Experiments on SANA and FLUX.1‑Krea show that DC‑Gen‑FLUX cuts 4K image generation latency by 53× on an NVIDIA H100 and, with NVFP4 SVDQuant, achieves a 138× total speedup on a single NVIDIA 5090 GPU.
arXiv:2606. 05981v2 Announce Type: replace-cross Abstract: Aggressive distillation of the diffusion U-Net inverts the per-frame bottleneck of real-time text-to-image pipelines: once the denoiser is a 4-step or 1-step distilled student, the text encoder becomes the critical path.
arXiv:2608. 04428v1 Announce Type: cross Abstract: Vision-language-action (VLA) models have emerged as a key component in embodied AI.
The paper investigates how attention sparsity behaves in autoregressive image generation, finding a distinct diagonal sparsity pattern due to spatial locality of visual tokens. It introduces a diagonal‑aware sparse attention mechanism that skips KV entries along the diagonal within a recent window, achieving up to 3.1× higher throughput and 1.19× lower latency with less than 2% quality loss compared to dense inference.
The paper introduces Abra, a family of flow‑matching transformers used to systematically study scaling laws for text‑to‑image diffusion models across three orders of magnitude in compute. It finds that diffusion models scale predictably like language models but need far more data, with compute optimality occurring at roughly 200 image tokens per parameter—ten times the optimal ratio for large language models. The study also shows that diffusion models are robust to overtraining, that more data is preferable to larger models, and that scaling predictability extends to generative quality, optimal CFG settings, representation quality, and training curve shapes.