SkyReels V4
Released Feb 25, 2026
Skywork AI's joint video-audio model (dual-stream MMDiT) that generates, inpaints and edits video from text, images, video clips, masks and audio references, at up to 1080p, 32 FPS and 15 seconds.
- Price per second
- $0.35
- Max clip length
- 15 s
- Max resolution
- 1080p
- Native audio
- Yes
- AA arena Elo
- 1103
- LMArena Elo
- —
- Capabilities
- text to video, image to video, video editing, extension
- Licence
- Proprietary
Strengths & weaknesses
Strengths
- Joint video and audio generation
- One model for text-to-video, image-to-video, extension and mask-guided editing
- Multimodal references: text, images, video, masks, audio
Weaknesses
- 15-second maximum
- Most expensive here at 1080p std: $0.35/s
- API 'fast' mode cannot generate audio yet; no open weights
releaseDate is arXiv v1 of the SkyReels-V4 paper (2602.21818, submitted 2026-02-25). No primary page gives a public API launch date; third-party sites suggest API access came in March 2026. Official API pricing is in credits (1 USD = 100 credits): without video input, V4 Std costs 35/14/11 credits per second and V4 Fast 27.5/11/8 at 1080P/720P/480P; with video input, Std costs 62.5/25/18 and Fast 50/20/15. The API's default is 1080p std. The API has no model-name string: V4 is called through endpoints such as https://api-gateway.skyreels.ai/api/v1/video/text2video/submit with mode 'std' or 'fast', so apiIds is empty. In the API, sound is an opt-in parameter (default False), and 'fast' mode does not support audio yet. No V4 weights on Hugging Face (Skywork has V2/V3 only) or GitHub (SkyworkAI/SkyReels-V3 is the latest), so openWeights false. skyreels-v4.ai is not verifiably official and was not used. Notable false as of 2026-10-27: AA-Video-T2V v2.1 #11 (Elo 1103, rank range 9–12, read from AA's Korean-language page) and not on LMArena's video board.