Seedance-2.0
#4 v Open-source jazykové modelybytedance · v2.0 · od Februar 2026 (China-Launch: ca. 7. Februar 2026; global/Dreamina: 24. Februar 2026) · 54× · naposledy 03. 7. 2026
70
Momentum
Seedance 2.0 is a proprietary, multimodal video generation model by ByteDance's SEED team, officially launched in China in early February 2026. It employs a Dual-Branch Diffusion Transformer architecture for simultaneous audio-video generation from text, image, audio, and video inputs. Native output is 480p/720p at up to 15 seconds; 1080p/2K is available via consumer platforms. The model is closed-source; no model weights are publicly released.
Vývoj momenta
04.04.03.07.
Vlastnosti
| Key Benchmark (%) | Artificial Analysis Video Arena T2V (with audio): Elo 1220, Rank 1 (Dreamina Seedance 2.0 720p); SeedVideoBench-2.0 I2V: Rank 1 on all 6 dimensions (Score 3.31–3.70/5) |
| Context Window (Tokens) | Approx. 308,880 tokens per 15-s video generation (per Volcengine documentation); no classic language-model context window |
| License | Proprietary / closed-source – no public model weights; commercial via subscriptions and API |
| Multimodality | Text + image (up to 9 refs) + video (up to 3 refs, ≤15 s) + audio (up to 3 refs, ≤15 s) as input; output: synchronized audio-video; 480p/720p native, 1080p/2K via platform; up to 15 s length |
| Platform | Jimeng/Jimeng AI (CN), Dreamina/CapCut (international), Volcengine API (CN), BytePlus/ModelArk (international), third-party: fal.ai, PiAPI, and others |
| Price | Volcengine API (official): 46 CNY/million tokens (regeneration) / 28 CNY/million tokens (with video input) ≈ 1 CNY/s (~$0.14/s); Dreamina from ~69 CNY/month (CN) / ~$18/month (int'l); BytePlus Lite API from ~$0.01/s |
| Price per 1M Tokens | 46 CNY (~$6.68) / million tokens (regeneration); 28 CNY (~$4.06) / million tokens (video-editing mode) – per Volcengine official pricing page |
| Release Date | February 2026 (China internal test from Feb. 7; Dreamina international from February 24, 2026) |