SSyncDev

SkyReels-V3-A2V-19B

SyncDev model directory · Profile updated Nov 18, 2026

Compare with another model

SkyReels-V3-A2V-19B is Skywork's open 720P talking-avatar model that animates a reference image from up to 200 seconds of driving audio.

Key specifications

Specification score
18.4
License
Skywork licence (other)

The specification score summarizes recorded features and limits. It is not a benchmark result.

Pricing conditions: No API pricing stated; open weights for self-hosting

Inputs and outputs

Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.

Accepts

  • Image input
  • Audio input
  • Text input

Produces

  • Video output

Capabilities

VisionAudioOpen weights

What is SkyReels-V3-A2V-19B?

Here is the constraint that decides most avatar projects: how long can the voice track be before the face falls apart? SkyReels-V3-A2V-19B accepts one audio track of up to 200 seconds, in mp3 or wav, and produces 720P video. For training clips, product explainers and dubbed presenters, that ceiling covers most real scripts in one pass.

It is the talking-avatar member of Skywork's SkyReels V3 family, which Skywork released with weights and inference code on 29 January 2026. The family is built on one multimodal in-context learning framework rather than three separate systems.

The V3 family at a glance

Model Job Size
SkyReels-V3-R2V-14B Video from one to four subject reference images 14B, 720P
SkyReels-V3-V2V-14B Video extension (single-shot 5 to 30 seconds) 14B, 720P
SkyReels-V3-A2V-19B Audio-driven talking avatar 19B, 720P

The A2V model is the largest, at about 18.9B parameters on disk.

Running it

Skywork recommends Python 3.12 or newer and CUDA 12.8 or newer. Single-GPU and torchrun multi-GPU commands are provided. The licence is Skywork's own, not a standard open-source licence, so read its terms before commercial use. An API for the V3 models launched on the same day through a third-party platform; check current availability and pricing there if you prefer not to host 19B parameters yourself.

What to watch in a pilot

  • Lip sync on fast speech. Test a script with rapid consonants and numbers.
  • Long-take stability. Generate close to the 200-second limit and watch the eyes and hairline in the final minute.
  • Language coverage. Try every language you plan to ship, not only English.
  • Identity drift. Compare the first and last frames against the reference image.

Where it fits

It belongs with other audio-driven systems in the audio-driven video directory. For game-style interactive video from the same lab, see Skywork's other releases. The Hub repository is dated 19 January, ten days before the announced release, which matters only if you are tracking revisions.

How to evaluate SkyReels-V3-A2V-19B

Compare the model on the work you need to complete.

Tasks and inputs

SkyReels-V3-A2V-19B is listed for Image-to-video, Audio-Driven Video. It accepts image, audio, text and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.

Capacity and cost

Choose duration, resolution and frame rate together. Compare the cost of an accepted clip, including repeat generations.

Choosing a deployment

The weights are downloadable under Skywork licence (other). Check runtime support and size the complete model and working context for your expected concurrency.

Frequently asked questions

One audio track of up to 200 seconds, in mp3 or wav format.

These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.

kyutai

AURA-1.3B

AURA-1.3B is Kyutai's real-time streaming talking-head model: a portrait plus live speech becomes 576p, 16 fps video, chunk by chunk, on one GPU.

— context

kyutai

AURA-5B

AURA-5B is Kyutai's larger streaming talking-head model: portrait plus speech to 720p, 25 fps video in four steps per chunk, at about 9.7 fps on one H100.

— context

Alibaba

Wan-Dancer-14B

Wan-Dancer-14B is Alibaba Wan's open music-to-dance video model, planning a whole track before refining it into minute-scale coherent dance.

— context

Meituan LongCat

LongCat Video Avatar 1.5

LongCat Video Avatar 1.5 generates audio-driven character video from speech, scene instructions and optional image references.

— context