Back to Gap Map Model components / Image, video, 3D & music generation

SkyReels

Skywork
closed / Overall score: 2.6

SkyReels is Skywork AI's line of video generation models. The current release, SkyReels V4, described in February 2026, generates video and synchronized audio together from text, image, video and audio prompts at up to 1080p and 15 seconds, and also handles inpainting and editing. Earlier generations, through SkyReels V3, were published with weights.

SkyReels V4 has no published weights. SkyReels V3 (January 2026; reference-to-video, video extension and talking-avatar models) and V2 (April 2025) remain downloadable under the Skywork Community License. The SkyReels-A1 and A2 portrait-animation models are a separate series and are not part of this entry.

Openness

1 medium confidence
1.0
weights
closed (no SkyReels V4 weights on Skywork's Hugging Face account, where a search for V4 returns nothing, or in its GitHub organization, which has no V4 repository)
data
closed (nothing is published for SkyReels V4 beyond its technical report)
code
closed (no code is published for SkyReels V4)
license
Proprietary (no license is published for SkyReels V4, which is available only as a hosted model)
prior-release
distributed (SkyReels V3 and V2: checkpoints download from the Hub without a gate under the Skywork Community License, which permits commercial use; inference code only)

SkyReels V4, Skywork's current release, has no published weights; Skywork describes it in a technical report and it is ranked on public arenas as a hosted model. SkyReels V3 and V2 remain downloadable under the Skywork Community License, which allows commercial use, so the open line trails the vendor's newest release.

Adoption

2 high confidence
2.0

Adoption is measured as Hugging Face downloads summed across the Skywork SkyReels V1 to V3 checkpoints, most of them V2. The current V4 has no downloadable weights to count.

Capability

3 medium confidence
3.0

SkyReels V4 places in the upper half of the public text-to-video arena, behind MiniMax H3, Wan 3.0 and Seedance, and higher on the image-to-video board. The open V3 and V2 models are not ranked there.

Verified 2026-11-12, except some scores, still awaiting re-confirmation