SkyReels V4: Video and Audio Generation, Inpainting, Editing and API Pricing

What SkyReels V4 is

SkyReels V4 is a video foundation model from Skywork AI that generates video and synchronised audio together and also handles inpainting and editing. Skywork AI described it in a technical report first submitted to arXiv on 25 February 2026, and the model is offered through the SkyReels API Platform and its Creative Studio. The report calls it the first video foundation model to combine multimodal input, joint video and audio generation, and a single treatment of generation, inpainting and editing. On LMArena's video leaderboard SkyReels V4 has 1459 Elo (rank 9).

Architecture

The model uses a dual-stream Multimodal Diffusion Transformer: one branch synthesises video and the other generates audio aligned in time with it, and both share a text encoder built on a multimodal large language model. The video branch uses in-context learning so that detailed visual guidance can be injected under complex conditioning, while the audio branch uses audio references to guide the sound it generates. On the video side, a channel-concatenation formulation puts image-to-video, video extension and video editing behind one interface, and extends to editing guided by reference images through multimodal prompts.

Output and inputs

SkyReels V4 supports output of up to 1080p at 32 frames per second and up to 15 seconds per clip, with multiple shots and synchronised audio. Inputs can include text, images, video clips, masks and audio references. To make high-resolution, long clips practical, Skywork AI generates a low-resolution version of the full sequence together with high-resolution keyframes, then applies dedicated super-resolution and frame-interpolation models. The API Platform lists Omni Reference to Video, text-to-video, image-to-video, talking avatar, video extension, video restyling and text editing among its features, along with automatic multi-shot generation.

API pricing

The SkyReels API is billed in credits bought in advance, at 100 credits per US dollar, and charges per second of video. Without video input, SkyReels V4 Std costs 35 credits per second at 1080p, 14 at 720p and 11 at 480p, which is $0.35, $0.14 and $0.11 per second. SkyReels V4 Fast costs 27.5, 11 and 8 credits per second at the same resolutions. With video input, Std costs 62.5, 25 and 18 credits per second and Fast 50, 20 and 15. A 10-second 1080p clip from V4 Std without video input therefore costs 350 credits, or $3.50.

Credit packages

Credits come in packages with a bonus: $30 buys 3,300 credits, $1,000 buys 115,000, $3,000 buys 360,000 and $5,000 buys 625,000, and larger amounts are priced on request. Skywork AI states that purchased credits are non-refundable, are valid for 12 months and then expire, and that the $30 first-time package can be used only once per account. The same pricing page lists the earlier SkyReels V3 avatar, lip-sync, extension, reference-to-video and restyling models, billed per second from 1 to 15 credits.

Official Resources