2x, 3x, or 4x upscale
Choose an integer scale (2, 3, 4) or a float in the range (1, 4]. Target resolution is aligned to the nearest 128-pixel boundary.
Video Upscale
Upscale video resolution up to 4x with a neural super-resolution model
You send a video URL and a scale factor (2, 3, or 4, or a float between 1 and 4). The service reads each frame, bicubic-upscales and center-crops to the nearest 128-pixel-aligned target resolution, then runs the frames through a neural video super-resolution model that adds real detail and sharpness. Original audio is muxed back into the output (copied without re-encoding). The video needs at least 17 frames. Output is MP4. Async: submit job, poll until completed, download the upscaled video.
videoprocessingstudio
Overview
Features
Neural super-resolution
Not simple upscaling — the model generates real detail, sharpness, and texture that bicubic interpolation cannot produce.
Audio preserved
Original audio is muxed back into the upscaled video (audio stream copied, no re-encoding).
Async processing
Submit job, poll until completed. Processing time depends on duration, resolution, and scale.
Use Cases
Archive restoration
Upscale old or low-resolution footage (e.g. 480p to 1920p) with added detail.
Display-ready delivery
Prepare content for 4K or large screens from HD sources.
Social media quality
Upscale clips before publishing for sharper appearance on high-DPI devices.
Input / Output
Input
Video URL and optional scale factor (default 4)
JSON body
Output
Upscaled video URL (async job result)
JSONmedia URL
Specs
- Latency
- Async; depends on duration, resolution, and scale
- Async
- true
- Rate Limit
- Per API key
- Max Input
- Minimum 17 frames. Target resolution must not exceed ~3.7M pixels (configurable).
Quickstart
Prerequisites
- -A CN8 Gateway API key with studio-upscale service enabled
- -A video URL (use upload-media to get one if needed)
1. Submit upscale job
studio-upscalePOST with the video URL and desired scale factor. The service reads frames, upscales with the super-resolution model, and muxes the original audio back.
POST/v1/proxy/studio-upscale
{
"videoUrl": "https://example.com/video.mp4",
"scale": 4
}Response
{
"status": "accepted",
"job_id": "job_upscale_abc",
"service": "studio-upscale"
}Poll the job endpoint until completed; then download the upscaled video from the output URL.
2. Get result
When the job is completed, the result contains the output video URL. The upscaled video has the original audio and increased resolution.
Services
View in API ReferenceVideo Upscale
POSTasyncUpscale video resolution using a neural super-resolution model. Scale 2x, 3x, or 4x (or float in (1,4]). Audio preserved. Async.
/v1/proxy/studio-upscale
Pricing
Billed per second of source video.
| Service | Unit | Price |
|---|---|---|
| Video Upscale | second | $0.08/second |
- -Price is per second regardless of scale factor.
Guides & Tips
What happens under the hood
- -Frames are read from the source video. Each frame is bicubic-upscaled to the scaled resolution, then center-cropped to the nearest 128-pixel-aligned target dimensions.
- -The frame sequence is padded to 8n+1 frames (model requirement). A neural video super-resolution model processes the sequence, generating real detail and texture.
- -Output frames are encoded as MP4. The original audio is muxed back without re-encoding.
Scale and resolution limits
- -Scale must be 2, 3, 4 (integer) or a float between 1 and 4.
- -Target resolution (after scale and 128-alignment) must not exceed ~3.7 million pixels (e.g. 1440x2560). This limit is configurable server-side.
- -If the target is too large, reduce the scale.
Frame requirement
- -The model requires at least 17 frames. Very short clips (under ~0.5s at 30fps) will be rejected.
- -Frames are padded internally to fit the model's 8n+1 requirement; this is handled automatically.
FAQ
Q: What scale factors are supported?
A: 2, 3, or 4 (integer), or a float in the range (1, 4]. Default is 4.
Q: Is the audio preserved?
A: Yes. The original audio track is copied into the upscaled output without re-encoding.
Q: What is the maximum output resolution?
A: Target resolution must not exceed approximately 3.7 million pixels (e.g. 1440x2560). Reduce scale if the source is already high-resolution.
Q: What is the minimum video length?
A: At least 17 frames (about 0.5 seconds at 30 fps). Shorter clips will be rejected.
Q: Does upscaling just make pixels bigger?
A: No. The neural model generates real detail, sharpness, and texture that simple interpolation cannot produce.
Related Products
Changelog
1.0 (2026-01-26)
- -Initial catalog.
1.1 (2026-02-23)
- -Aligned with upscale_service: neural super-resolution, scale 2/3/4 or float (1,4], 128-aligned target, 17-frame minimum, ~3.7M pixel limit, audio mux. No model name in docs.