Video Upscale Models - Parameters Reference
Scenario offers three main models for video upscaling, each optimised for different tasks. These models increase the resolution of videos, enhance sharpness and remove artifacts. When calling /generate/custom/{modelId} with a video upscaling model, include the appropriate parameters as described below. Note that video upscaling typically operates on short clips (under around 40 seconds) due to processing costs and memory requirements.
Topaz Video Upscale
Model ID: model_topazlabs-video-upscale
Model purpose: Topaz Labs’ industry‑leading video upscaler improves low‑resolution or choppy videos by sharpening details, reducing artifacts and boosting resolution up to 4K. It can also increase frame rates up to 60 fps, producing smoother motion.
- Supported input length: Typically suited for short clips; processing long videos may require splitting into segments.
- Output: Up to 4 K resolution; optional frame‑rate interpolation to 60 fps.
| Parameter | Type | Typical values | Description |
|---|---|---|---|
video | assetId | – | Video to upscale. Required. |
resolution | number | ["720p", "1080p", "4k"] | Target resolution |
fps | number | [15,60] | Target Frame-Per-Second |
Runway Upscale v1
Model ID: model_runway-upscale-v1
Model purpose: A simple and efficient video upscaler from Runway. It enlarges video frames by up to four times and produces crisp 4 K output. The model maintains the original frame rate and is best for short clips (typically under ~40 seconds).
- Supported input length: Recommended for videos under about 40 seconds.
- Output: 4× upscaling to 4K resolution; frame rate remains unchanged.
| Parameter | Type | Typical values | Description |
|---|---|---|---|
video | assetId | – | Video file to upscale. Videos must be shorter than 40s and less than 4096px per side. |
Seed VR2 Upscale Video
Model ID: model_seedvr-2-upscale-video
Model purpose: Based on the Seed VR2 diffusion transformer, this model performs both restoration and upscaling in a single step. It removes blur, reduces noise and recovers fine details. It maintains temporal consistency across frames and supports arbitrary input sizes up to 4K.
- Supported input length: Suitable for short clips; for long videos split into segments.
- Scaling: Supports both factor‑based and target‑resolution modes; factors range from 2× to 10×. limited to 4K.
- Output: Up to 2160 p (4 K) depending on chosen
targetResolution.
| Parameter | Type | Typical values | Description |
|---|---|---|---|
videoUrl | assetId | – | The video to upscale. Required. |
upscaleMode | string (enum) | factor, target | Selects whether to use an explicit factor (factor) or target resolution (target). |
upscaleFactor | number | 2–10 | Scaling factor when upscaleMode is factor. |
targetResolution | string (enum) | 720p, 1080p, 1440p, 2160p | Target output resolution when upscaleMode is target. |
noiseScale | number | 0–1 | Amount of noise introduced during diffusion; lower values preserve more original detail. |
outputQuality | string (enum) | low, medium, high, maximum | Controls encoding quality; higher values produce larger files. |
outputWriteMode | string (enum) | fast, balanced, small | Chooses between faster encoding, a balance between speed and size, or smaller file size. |
seed | number | – | Optional random seed for reproducibility. |
Notes:
- Video upscaling models often have strict clip length and resolution limits due to processing costs. If you need to upscale a long video, consider splitting it into shorter segments and processing each separately.
- The parameters shown above reflect typical usage. Additional parameters may be added or modified over time; refer to Scenario’s API documentation for the latest information.
Updated 2 days ago