Seedream
ByteDance
Seedream is an image generation model line from ByteDance. The latest release is Seedream 5.0 Lite (13 Feb 2026). This page tracks all 5 Seedream releases since 1 Dec 2024, with specs, licensing, and how to access each one. Official site →
Seedream 5.0 Lite
Adds deep thinking reasoning and real-time web search to image generation
- License
- Proprietary
- Commercial use
- Allowed
- Availability
- Official API, Consumer app
Seedream 4.5
Refinement of 4.0 focused on editing fidelity: cross-image consistency module for stable character/clothing/lighting across a batch; up to 14 reference images; sharper small-text and face rendering; 1K output dropped
- License
- Proprietary
- Commercial use
- Allowed
- Availability
- Official API, Consumer app
- Resolution
- 2K-4K
Released via Volcengine/BytePlus as Doubao-Seedream-4.5. Public test window 2025-12-03 to 2025-12-09, full release from 2025-12-10 (Beijing-time reports say early 04:00 on 2025-12-04). No single ByteDance Seed blog post fetched for the exact day - date rests on the Volcengine rollout schedule and consistent secondary coverage. Closed, API-only plus Doubao/Jimeng.
Seedream 4.0
Unified T2I, image editing, and multi-image composition in one model; highly efficient DiT with powerful VAE reducing image tokens; 1.4s for 2K image generation; multimodal post-training via fine-tuned VLM
- License
- Proprietary
- Commercial use
- Allowed
- Availability
- Official API, Consumer app
- Resolution
- 1K-4K
Date approximate - paper states 'integrated into Doubao and Jimeng as of September 2025'. Accessible on Volcano Engine API as Doubao-Seedream-4.0.
Seedream 3.0
Improved prompt alignment for complex instructions; fine-grained typography; 4-8x inference speedup via consistent noise expectation; doubled training dataset with defect-aware paradigm
- License
- Proprietary
- Commercial use
- Allowed
- Availability
- Official API, Consumer app
Date approximate - technical report states 'integrated into multiple platforms in early April 2025'.
Seedream 2.0
Native Chinese-English bilingual image generation; self-developed bilingual LLM as text encoder; strong cultural knowledge and text rendering in both languages
- License
- Proprietary
- Commercial use
- Allowed
- Availability
- Consumer app
Date approximate - paper states 'incorporated into Doubao and Dreamina as of early December 2024'. Available via Doubao (豆包) and Jimeng (即梦/Dreamina).
See Seedream in context on the image model timeline, or browse all model families.