Tasks and inputs
SkyReels-V3-R2V-14B is listed for Image-text-to-video, Image-to-video. It accepts text, image and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 19, 2026
SkyReels-V3-R2V-14B is Skywork's 14B reference-to-video model that turns one to four reference images and a prompt into 720p clips with consistent subjects.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Open weights under the Skywork License; no first-party per-clip price listed. A partner-hosted API exists, so confirm pricing on your own account.
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
Produces
Keeping the same face, product or set across a generated clip is where most video models slip. SkyReels-V3-R2V-14B is Skywork's model built for exactly that: give it between one and four reference images (characters, objects, backgrounds) and a text prompt, and it composes a video that preserves them. Skywork released weights and inference code on 29 January 2026, alongside the video-extension and talking-avatar models of the SkyReels V3 family.
| Repository | Skywork/SkyReels-V3-R2V-14B |
| Parameters | ~14.3B |
| References | 1-4 images |
| Recommended output | 5 seconds, 720p, 24 fps |
| Aspect ratios | 1:1, 3:4, 4:3, 16:9, 9:16 |
| Licence | Skywork License |
A common failure in reference-driven video is pasting the reference into the frame unchanged. Skywork's data pipeline pairs frames across different moments of the same video, uses image-editing models to extract subjects and complete backgrounds, and rewrites captions semantically, so the model learns to re-render subjects rather than stamp them. Training mixes images with video and multiple resolutions at once.
Skywork publishes this comparison:
| Model | Reference consistency | Instruction following | Visual quality |
|---|---|---|---|
| Vidu Q2 | 0.5961 | 27.84 | 0.7877 |
| Kling 1.6 | 0.6630 | 29.23 | 0.8034 |
| PixVerse V5 | 0.6542 | 29.34 | 0.7976 |
| SkyReels V3 | 0.6698 | 27.22 | 0.8119 |
SkyReels leads on reference consistency and visual quality. On instruction following it is actually the lowest of the four, even though Skywork highlights its score. If prompt adherence matters more than likeness for your use, weigh that.
Inference uses a single script with --task_type reference_to_video and --ref_imgs as a comma-separated list. On GPUs under 24 GB, add --low_vram for FP8 weight-only quantisation with block offload, and drop --resolution to 540P or 480P. Multi-GPU runs use xDiT USP. Skywork also lists a hosted API through a partner platform, but no first-party price is published here.
Use your real assets: three product shots and a person, say, then ask for scenes that move the camera around them. Score identity drift at the last frame, not the first. For the other SkyReels V3 tasks, see SkyReels-V3-V2V-14B for extension and SkyReels-V3-A2V-19B for talking avatars. Commercial alternatives include Kling VIDEO 4.0; more are under image-to-video.
Compare the model on the work you need to complete.
SkyReels-V3-R2V-14B is listed for Image-text-to-video, Image-to-video. It accepts text, image and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.
Choose duration, resolution and frame rate together. Compare the cost of an accepted clip, including repeat generations.
The weights are downloadable under Skywork License. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
tbc
TBC mischief 1 is an image-to-video model built on Wan 2.2, sold on fal and OpenRouter at about half a cent per second for 3-5 s clips at up to 720p.
NVIDIA
Cosmos3-H-Dreams is NVIDIA's open real-time surgical world model: it streams Versius robot video from one endoscope frame and two-arm commands.
NVIDIA
Cosmos3-H-Embodied-Nano is NVIDIA's open 16B surgical world model that does forward dynamics, inverse dynamics and policy from one checkpoint.
NVIDIA
Cosmos3-Nano-Sim-Bimanual is NVIDIA's open 16B world model that simulates video of a two-armed AgiBot robot as it moves its camera and manipulates.
noiz-ai
Helix World (preview) is an open real-time world model from Noiz AI and HKUST that streams explorable 768x512 video with stereo sound that follows the camera.
Alibaba
Wan-Animate-2 is Alibaba Wan's open 14B character-animation model that drives a character from video directly, with text-controlled camera angle.