SkyReels API for Python
Python client for SkyReels-class long-form text-to-video and image-to-video through a hosted API on Synexa
Run SkyReels in the browser → View on GitHub
Install
pip install git+https://github.com/skyreels-dev/skyreels-api.git
export SYNEXA_API_KEY="sk-..."Quickstart
import skyreels_api
output = skyreels_api.run({
"prompt": "A cinematic shot of a lighthouse at dawn, soft fog, warm light"
})
print(output)Hosted models
- bytedance/seedance-2.5 — Seedance 2.5 generates a single-shot video of up to 30 seconds from a text prompt, with synchronised audio. ($0.473 / run)
- tongyi/wan2.2 — Generate 5s 480p videos using Wan 2.2 14B. A comprehensive video foundation models that pushes the boundaries of video generation. ($0.2 / run)
About SkyReels
SkyReels-V2 is Skywork AI's open video foundation model for long clips, using diffusion forcing to extend a shot autoregressively past the usual five-second limit, with 1.3B, 5B and 14B checkpoints at 540p and 720p. This Python client gives you the same long-form video generation as a hosted API call, using Synexa's bytedance/seedance-2.5 endpoint for up to 30 second shots with audio, plus Wan 2.2 image-to-video. One pip install, a blocking run() call, polling and webhooks, and per-run pricing.
FAQ
Is there a SkyReels API?
Skywork publishes SkyReels-V2 as open weights and a PyTorch repository, not as a public REST API. This package wraps a hosted Synexa endpoint that provides the same long-form video generation capability, so you can call it over HTTPS without running the model.
How much does the SkyReels API cost?
The default endpoint, bytedance/seedance-2.5, is $0.473 per run, and that covers any length from 4 to 30 seconds with or without audio. The tongyi/wan2.2 image-to-video endpoint is $0.20 per 5 second clip. Billing is per completed run.
Can I run SkyReels without a GPU?
Not locally; even the 1.3B model needs a CUDA GPU and the 14B models need a large one. With this client the generation happens on the hosted side, so a laptop or a serverless function is enough.
Does this client work with the original SkyworkAI SkyReels-V2 repo or ComfyUI?
No. It does not load local checkpoints, run diffusion-forcing inference, or talk to the ComfyUI SkyReels nodes. It is an HTTP client for the hosted endpoints only.
What input formats does it accept?
For bytedance/seedance-2.5 only prompt is required; resolution (480p, 720p, 1080p), duration (4 to 30 seconds or auto), aspect_ratio (16:9, 9:16, 1:1, 21:9 or auto), generate_audio and bitrate_mode are optional. For tongyi/wan2.2 you pass a prompt and an input_image, with optional aspect_ratio, num_frames and seed.
Is this the official SkyReels SDK?
No. This is an independent client and is not affiliated with Skywork AI. The official project is at https://github.com/SkyworkAI/SkyReels-V2.