ByteDance’s vCube video upscaler reconstructs detail and improves visual quality in low-resolution video. You can take a 480p, 720p, or 1080p clip and push it up to 2K or 4K at frame rates up to 60fps. It’s built for the kind of content people are generating today — AI video, short-form drama, user-generated clips — and works well on real footage too.

The model pairs nicely with bytedance/seedance-2.0, which generates great 720p or 1080p video that you can then upscale to 4K. But you can feed it any video.

Inputs

• video — The video to upscale. Any common format works (mp4, mov, etc.).
• processing_type — standard is the default and works for everything. pro is a higher-quality model with materially better skin texture and fine detail — great for real-people footage. Pro is allowlist-only and is currently enabled on the Replicate account.
• scene — Tell the model what kind of content this is so it picks the right enhancement preset:
• aigc — AI-generated video (recommended for Seedance outputs)
• short_series — short-form drama and episodic content
• ugc — user-generated content (phone footage, social media)
• old_film — film restoration
• common — anything else
• target_resolution — Output resolution. Choose from 240p, 360p, 480p, 540p, 720p, 1080p, 2k, or 4k. 4K is the native maximum.
• target_fps — Output frame rate. Choose from 24, 30, 60, or 120. Frame interpolation is applied if the target is higher than the source.

Output

A single mp4 file at the requested resolution and frame rate.

Tips

• For AI-generated content like Seedance outputs, use scene=aigc. It’s tuned for the textures and motion you get from generative video models.
• Use the pro tier when faces and skin are in frame. The quality difference on real-people footage is significant.
• Going from 720p to 4K at 60fps is the most common path. The model does both spatial upscaling and frame interpolation in one pass.
• A 5-second clip typically takes 3–5 minutes end-to-end, depending on the tier and the source resolution.

Pricing

Billed per second of output video. Cost depends on the processing tier (standard vs pro) and the target resolution and frame rate. See the pricing tab on this page for the full breakdown.