mlsanityData to 14 Oct 2026

Seedream 4.0: Toward Next-generation Multimodal Image Generation

arXiv 2509.20427 ↗First posted 24 Sep 2025cs.CVcited by 249

On the map: Efficient LLM inference and serving → Visual token pruning in multimodal models

Abstract (arXiv)

We introduce Seedream 4.0, an efficient and high-performance multimodal image generation system that unifies text-to-image (T2I) synthesis, image editing, and multi-image composition within a single framework. We develop a highly efficient diffusion transformer with a powerful VAE which also can reduce the number of image tokens considerably. This allows for efficient training of our model, and enables it to fast generate native high-resolution images (e.g., 1K-4K). Seedream 4.0 is pretrained on billions of text-image pairs spanning diverse taxonomies and knowledge-centric concepts. Comprehensive data collection across hundreds of vertical scenarios, coupled with optimized strategies, ensures stable and large-scale training, with strong generalization. By incorporating a carefully fine-tuned VLM model, we perform multi-modal post-training for training both T2I and image editing tasks jointly. For inference acceleration, we integrate adversarial distillation, distribution matching, and quantization, as well as speculative decoding. It achieves an inference time of up to 1.8 seconds for generating a 2K image (without a LLM/VLM as PE model). Comprehensive evaluations reveal that Seedream 4.0 can achieve state-of-the-art results on both T2I and multimodal image editing. In particular, it demonstrates exceptional multimodal capabilities in complex tasks, including precise image editing and in-context reasoning, and also allows for multi-image reference, and can generate multiple output images. This extends traditional T2I systems into an more interactive and multidimensional creative tool, pushing the boundary of generative AI for both creativity and professional applications. We further scale our model and data as Seedream 4.5. Seedream 4.0 and Seedream 4.5 are accessible on Volcano Engine https://www.volcengine.com/experience/ark?launch=seedream.

Key facts. Cited by 249 papers in the map (data through 14 Oct 2026). Placed in Visual token pruning in multimodal models, one of 12 directions in Efficient LLM inference and serving. Builds on 18 of its 20 references that are in the map. First posted on arXiv 24 Sep 2025.

Builds on

18 of 20 references.

References · 18

  1. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  2. 2023Adding Conditional Control to Text-to-Image Diffusion Modelscited by 5,503
  3. 2023ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generationcited by 1,045
  4. 2025FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Spacecited by 1,006
  5. 2025Flow-GRPO: Training Flow Matching Models via Online RLcited by 661
  6. 2025DanceGRPO: Unleashing GRPO on Visual Generationcited by 413
  7. 2025Seed1.5-VL Technical Reportcited by 373
  8. 2024Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesiscited by 156
  9. 2024ControlNet++: Improving Conditional Controls with Efficient Consistency Feedbackcited by 154
  10. 2025Seedream 3.0 Technical Reportcited by 154
  11. 2025Diffusion Adversarial Post-Training for One-Step Video Generationcited by 151
  12. 2025AdaCoT: Pareto-Optimal Adaptive Chain-of-Thought Triggering via Reinforcement Learningcited by 87
6 more
  1. 2025RewardDance: Reward Scaling in Visual Generationcited by 68
  2. 2025Seedream 2.0: A Native Chinese-English Bilingual Image Generation Foundation Modelcited by 62
  3. 2024SeedEdit: Align Image Re-Generation to Image Editingcited by 61
  4. 2025Adversarial Distribution Matching for Diffusion Distillation Towards Efficient Image and Video Synthesiscited by 41
  5. 2025Hyper-Bagel: A Unified Acceleration Framework for Multimodal Understanding and Generationcited by 13
  6. 2025RayFlow: Instance-Aware Diffusion Acceleration via Adaptive Flow Trajectoriescited by 12

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 52; 2026: 197.

249 papers.

Sort:

By area:

  1. 2025Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformercited by 277
  2. 2026Seedance 2.0: Advancing Video Generation for World Complexitycited by 160
  3. 2025HunyuanImage 3.0 Technical Reportcited by 149
  4. 2025LongCat-Image Technical Reportcited by 100
  5. 2025Seedance 1.5 pro: A Native Audio-Visual Joint Generation Foundation Modelcited by 80
  6. 2026Being-H0.5: Scaling Human-Centric Robot Learning for Cross-Embodiment Generalizationcited by 66
  7. 2025Pico-Banana-400K: A Large-Scale Dataset for Text-Guided Image Editingcited by 61
  8. 2025BLIP3o-NEXT: Next Frontier of Native Image Generationcited by 54
  9. 2025Distribution Matching Distillation Meets Reinforcement Learningcited by 50
  10. 2025Thinking with Video: Video Generation as a Promising Multimodal Reasoning Paradigmcited by 42
  11. 2026FireRed-Image-Edit-1.0 Technical Reportcited by 36
  12. 2026Qwen-Image-2.0 Technical Reportcited by 36
38 more
  1. 2026HiDream-O1-Image: A Natively Unified Image Generative Foundation Model with Pixel-level Unified Transformercited by 32
  2. 2026SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecturecited by 32
  3. 2026Gen-Searcher: Reinforcing Agentic Search for Image Generationcited by 30
  4. 2026InternVL-U: Democratizing Unified Multimodal Models for Understanding, Reasoning, Generation and Editingcited by 28
  5. 2026VINO: A Unified Visual Generator with Interleaved OmniModal Contextcited by 26
  6. 2025UniGenBench++: A Unified Semantic Evaluation Benchmark for Text-to-Image Generationcited by 25
  7. 2025Phased DMD: Few-step Distribution Matching Distillation via Score Matching within Subintervalscited by 24
  8. 2026D-OPSD: On-Policy Self-Distillation for Continuously Tuning Step-Distilled Diffusion Modelscited by 22
  9. 2026GEMS: Agent-Native Multimodal Generation with Memory and Skillscited by 21
  10. 2025Factuality Matters: When Image Generation and Editing Meet Structured Visualscited by 20
  11. 2026Visual Generation Unlocks Human-Like Reasoning through Multimodal World Modelscited by 19
  12. 2025ImAgent: A Unified Multimodal Agent Framework for Test-Time Scalable Image Generationcited by 17
  13. 2025MIRA: Multimodal Iterative Reasoning Agent for Image Editingcited by 17
  14. 2026Unify-Agent: A Unified Multimodal Agent for World-Grounded Image Synthesiscited by 17
  15. 2025PICABench: How Far Are We from Physically Realistic Image Editing?cited by 16
  16. 2026Wan-Image: Pushing the Boundaries of Generative Visual Intelligencecited by 16
  17. 2025MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMscited by 15
  18. 2026Think-Then-Generate: Reasoning-Aware Text-to-Image Diffusion with LLM Encoderscited by 15
  19. 2025Flow Map Distillation Without Datacited by 14
  20. 2026DeepGen 1.0: A Lightweight Unified Multimodal Model for Advancing Image Generation and Editingcited by 14
  21. 2026GEditBench v2: A Human-Aligned Benchmark for General Image Editingcited by 14
  22. 2025SCAIL: Towards Studio-Grade Character Animation via In-Context Learning of 3D-Consistent Pose Representationscited by 13
  23. 2025EasyV2V: A High-quality Instruction-based Video Editing Frameworkcited by 13
  24. 2025Adversarial Flow Modelscited by 12
  25. 2026How Well Do Models Follow Visual Instructions? VIBE: A Systematic Benchmark for Visual Instruction-Driven Image Editingcited by 12
  26. 2026GENIUS: Generative Fluid Intelligence Evaluation Suitecited by 12
  27. 2026JoyAI-Image: Awaking Spatial Intelligence in Unified Multimodal Understanding and Generationcited by 12
  28. 2026Lance: Unified Multimodal Modeling by Multi-Task Synergycited by 12
  29. 2025WiseEdit: Benchmarking Cognition- and Creativity-Informed Image Editingcited by 11
  30. 2026ERNIE 5.0 Technical Reportcited by 11
  31. 2026BizGenEval: A Systematic Benchmark for Commercial Visual Content Generationcited by 11
  32. 2026GenClaw: Code-Driven Agentic Image Generationcited by 11
  33. 2025UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Editscited by 10
  34. 2025MICo-150K: A Comprehensive Dataset Advancing Multi-Image Compositioncited by 10
  35. 2026SpatialEdit: Benchmarking Fine-Grained Image Spatial Editingcited by 10
  36. 2026PosterOmni: Generalized Artistic Poster Creation via Task Distillation and Unified Reward Feedbackcited by 9
  37. 2026RealRestorer: Towards Generalizable Real-World Image Restoration with Large-Scale Image Editing Modelscited by 9
  38. 2026CreatiParser: Generative Image Parsing of Raster Graphic Designs into Editable Layerscited by 9

50 most cited of 249 All 249 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025ShareGPT-4o-Image: Aligning Multimodal Models with GPT-4o-Level Image Generation
  2. 2026Unified Text-Image Generation with Weakness-Targeted Post-Training
  3. 2025Muddit: Liberating Generation Beyond Text-to-Image with a Unified Discrete Diffusion Model
  4. 2026LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes
  5. 2025SeedEdit 3.0: Fast and High-Quality Generative Image Editing
  6. 2025Query-Kontext: An Unified Multimodal Model for Image Generation and Editing
  7. 2025An Empirical Study of GPT-4o Image Generation Capabilities
  8. 2026NextFlow: Unified Sequential Modeling Activates Multimodal Understanding and Generation