Tagged articles
diffusion transformer
14 articles · Page 1 of 1
SmartPhotoCrafter introduces a unified reasoning-to-generation framework for automatic photo enhancement: an Image Critic analyzes quality via chain-of-thought reasoning, then a Photographic Artist performs high-fidelity restoration and aesthetic retouching, unifying dehazing, deblurring, and color grading without explicit user instructions, trained via a three-stage strategy with multi-level rewards.
SmartPhotoCrafterautomatic photo enhancementchain-of-thought
0 likes · 13 min read

LLaDA-Image is a 6B diffusion transformer that learns visual priors from images alone — 90% of its 220M training samples use image-only supervision — then aligns language in later fine-tuning, achieving top open-source scores on Qwen-Image-Bench, supporting both text-to-image and instruction-based editing, and distilling to 2–4 steps via TwinFlow.
Image-Only PretrainingLLaDA-ImageQwen-Image-Bench
0 likes · 21 min read

Atlas, a multimodal autoregressive diffusion Transformer introduced by Fei‑Fei Li's World Labs, can generate pixel‑level camera‑controlled images and videos, perform high‑quality 3D reconstruction from a few photos, simulate space‑time for robotics, and outperforms state‑of‑the‑art models in quantitative evaluations.
3D ReconstructionAtlasRobotics
0 likes · 8 min read


21CTOSep 2, 2026 · Artificial Intelligence
World Labs, founded by Fei‑Fei Li and colleagues, launched Atlas—a multimodal autoregressive diffusion transformer that natively handles text, images, video and 3D, offering camera‑control generation, spatial reconstruction, spatiotemporal simulation and image synthesis, and achieving leading scores on human preference and 3D reconstruction benchmarks.
3D ReconstructionAIAtlas
0 likes · 6 min read

TF-CoDiT introduces a diffusion‑Transformer framework that converts multi‑channel treasury futures time series into discrete wavelet coefficients, encodes cross‑channel dependencies with a U‑shaped VAE, conditions generation on a structured FinMAP prompt, and achieves state‑of‑the‑art MSE and MAE scores across multiple contracts and horizons.
FinMAPTF-CoDiTU-VAE
0 likes · 17 min read

ProMoE introduces a two‑step routing MoE framework with explicit semantic guidance that tackles the high spatial redundancy and functional heterogeneity of visual tokens, enabling diffusion transformers to scale efficiently and outperform dense models and prior MoE approaches across generation, convergence, and scaling benchmarks.
Explicit RoutingPrototypical RoutingRouting Contrastive Learning
0 likes · 9 min read

Alibaba's Tongyi releases the Z-Image base model, a non‑distilled diffusion transformer that supports full classifier‑free guidance, negative prompts, higher diversity, and fine‑tuning, contrasting with the faster Turbo variant and providing detailed usage instructions and community resources.
AlibabaClassifier-Free GuidanceImage Generation
0 likes · 4 min read


21CTOOct 20, 2025 · Artificial Intelligence
World Labs unveiled RTFM, a real‑time frame model that runs on a single H100 GPU, generating persistent, interactive 3D worlds from 2D images without explicit 3D representations, highlighting the growing computational demands of generative world models and their potential to reshape AI-driven spatial intelligence.
3D generationGPU accelerationGenerative AI
0 likes · 9 min read

The Bilibili TTV team optimized OpenSora and CogVideoX text‑to‑video models by redesigning data storage with Alluxio, parallelizing VAE encoding, applying dynamic sequence‑parallel and DeepSpeed‑Ulysses attention, adapting GPU code for NPU execution, leveraging profiling‑driven kernel fusion, FlashAttention, and expandable memory to dramatically increase training efficiency and frame throughput, while outlining future pipeline‑parallel and ZeRO‑3 scaling plans.
FlashAttentionNPUdata pipeline
0 likes · 26 min read


AIWalkerJan 15, 2025 · Artificial Intelligence
Magic Mirror introduces a single‑stage, zero‑shot framework that fuses dual facial embeddings with a conditional adaptive normalization module inside a Video Diffusion Transformer, achieving superior identity consistency, natural dynamics, and high visual quality compared with existing video generation methods.
conditional adaptive normalizationdiffusion transformeridentity preservation
0 likes · 16 min read

EasyAnimate v3, an open‑source video generation system from Alibaba Cloud AI Platform, introduces Diffusion Transformer‑based architecture, Hybrid Motion Module, and Slice VAE to enable image‑to‑video, text‑to‑video, and unlimited‑length video creation with up to 720p/144 fps resolution on modest GPU memory.
AIComputer VisionEasyAnimate
0 likes · 5 min read

EasyAnimate, an open‑source DiT‑based video generation framework from Alibaba Cloud AI Platform PAI, offers a complete pipeline—including data preprocessing, VAE and DiT training, LoRA fine‑tuning, motion‑module integration, and scalable inference up to 768×768 resolution and 144 frames—leveraging Diffusion Transformers to produce longer, higher‑quality videos.
AI videoLoRAVAE
0 likes · 14 min read

Tencent has open‑sourced its upgraded 1.5‑billion‑parameter HunYuan DiT model—the first Chinese‑native, bilingual (Chinese‑English) text‑to‑image diffusion‑with‑transformer system—delivering about 20% visual quality improvement, multi‑round generation, video‑generation potential, and free commercial use, with full weights, inference code, and algorithms available on Hugging Face and GitHub for developers and enterprises.
Chinese-native AIDiT architectureMultimodal Generation
0 likes · 6 min read


NewBeeNLPMar 20, 2024 · Artificial Intelligence
This article provides a comprehensive technical walkthrough of Open‑Sora 1.0, covering its Diffusion‑Transformer architecture, three‑stage training strategy, data‑preprocessing scripts, generation quality, and the Colossal‑AI acceleration that together make Sora‑level video synthesis openly reproducible.
AI videoOpen-Soradiffusion transformer
0 likes · 12 min read
