Seedream 3.0 Technical Report

2025-04-15

AI Summary

Seedream 3.0 addresses key limitations of Seedream 2.0—including inadequate complex prompt alignment, suboptimal Chinese typography rendering, insufficient visual aesthetic modeling, constrained image fidelity, and fixed 1K-resolution output—by introducing the first bilingual (Chinese–English) foundational image generation model natively supporting 2K-resolution output. Methodologically, it proposes a novel defect-aware training paradigm and a dual-axis collaborative sampling framework, incorporating a consistent noise expectation acceleration mechanism. It further integrates hybrid-resolution training, cross-modal RoPE positional encoding, representation alignment loss, resolution-aware timestep sampling, vision-language model (VLM)-based reward modeling, and supervised fine-tuning (SFT) with aesthetic annotations. Experiments demonstrate substantial improvements: significantly enhanced Chinese text rendering quality, markedly improved alignment with human aesthetic preferences, state-of-the-art image fidelity and typographic professionalism, and 4–8× faster inference speed.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

We present Seedream 3.0, a high-performance Chinese-English bilingual image generation foundation model. We develop several technical improvements to address existing challenges in Seedream 2.0, including alignment with complicated prompts, fine-grained typography generation, suboptimal visual aesthetics and fidelity, and limited image resolutions. Specifically, the advancements of Seedream 3.0 stem from improvements across the entire pipeline, from data construction to model deployment. At the data stratum, we double the dataset using a defect-aware training paradigm and a dual-axis collaborative data-sampling framework. Furthermore, we adopt several effective techniques such as mixed-resolution training, cross-modality RoPE, representation alignment loss, and resolution-aware timestep sampling in the pre-training phase. During the post-training stage, we utilize diversified aesthetic captions in SFT, and a VLM-based reward model with scaling, thereby achieving outputs that well align with human preferences. Furthermore, Seedream 3.0 pioneers a novel acceleration paradigm. By employing consistent noise expectation and importance-aware timestep sampling, we achieve a 4 to 8 times speedup while maintaining image quality. Seedream 3.0 demonstrates significant improvements over Seedream 2.0: it enhances overall capabilities, in particular for text-rendering in complicated Chinese characters which is important to professional typography generation. In addition, it provides native high-resolution output (up to 2K), allowing it to generate images with high visual quality.

Problem

• Improves alignment with complex prompts in image generation
• Enhances fine-grained typography and text-rendering capabilities
• Increases image resolution and visual fidelity significantly

Innovation

• Defect-aware training and dual-axis data sampling
• Mixed-resolution training and cross-modality RoPE
• Consistent noise expectation for speedup