Supported Upscaling Models
WaveSpeed's video upscaling models are designed for different quality targets, throughput requirements, and budgets. Both are available through the same API authentication.
SeedVR2 Video
Premium option for high-fidelity upscaling with temporal consistency across frames. Keeps reconstructed detail stable through motion.
Best for: cinematic content, AI-generated video, motion-heavy footage
Standard Video Upscaler
Faster and more cost-effective. Improves clarity, restores detail, and supports output up to 4K for general-purpose enhancement.
Best for: batch pipelines, user-uploaded content, cost-sensitive workflows
| SeedVR2 Video | Standard Video Upscaler | |
|---|---|---|
| Temporal consistency | Yes, frame-aware | Yes, optimized |
| Processing speed | ~10 to 30s per second of video | ~5 to 10s per second of video |
| Cost at 1080p | $0.15 per 5 seconds | $0.025 per 5 seconds |
| Cost at 4K | $0.25 per 5 seconds | $0.10 per 5 seconds |
| Best use | Cinematic, motion-heavy, AI-generated | Batch, general-purpose, cost-sensitive |
If you want the highest fidelity, SeedVR2 is the better fit. For routine enhancement at scale, the standard upscaler is the practical starting point.
Upscale Video via API
How It Works
Submit
Send a video URL with your target resolution (720p, 1080p, 2K, or 4K)
Process
The model runs inference across frames, rebuilding detail and reducing artifacts
Retrieve
Poll for the job result or configure a webhook to receive the output URL automatically
curl --location --request POST \
"https://api.wavespeed.ai/api/v3/wavespeed-ai/video-upscaler" \
--header "Content-Type: application/json" \
--header "Authorization: Bearer ${WAVESPEED_API_KEY}" \
--data-raw '{
"video": "https://example.com/input.mp4",
"target_resolution": "1080p",
"copy_audio": true
}'To route to SeedVR2 specifically, use the endpoint /api/v3/wavespeed-ai/seedvr2-video. For full parameter reference, see the WaveSpeed API documentation.
Pricing
Per-second billing with a 5-second minimum. No seat fees, no subscription.
| Model | 720p | 1080p | 2K | 4K |
|---|---|---|---|---|
| Standard Video Upscaler | $0.025 | $0.025 | $0.05 | $0.10 |
| SeedVR2 Video | $0.10 | $0.15 | $0.20 | $0.25 |
Prices per 5 seconds. Verify current pricing at wavespeed.ai/models.
FAQ
What is the difference between SeedVR2 and the standard video upscaler?
SeedVR2 processes frames with temporal awareness, so it knows what came before and after each frame. That keeps motion more stable and textures more consistent across the full clip. The standard upscaler runs faster and costs significantly less per second, making it the practical choice for batch work where peak motion quality is not the priority.
What resolutions does WaveSpeed support?
720p, 1080p, 2K, and 4K. Set target_resolution in your API request. Both models support all four options.
Does AI upscaling work on AI-generated video?
Yes. Diffusion models typically output at 480p or 720p with slightly unstable motion. SeedVR2's temporal logic is designed to help stabilize that while upscaling, so the result tends to be cleaner than you might expect from synthetic source material.
Is audio preserved during processing?
Set copy_audio: true and the audio track passes through untouched. Timing, sync, and encoding stay exactly as they were in the source file.
What is the maximum video length per job?
10 minutes. For longer content, split into segments, process in parallel, and merge afterward.
Do I need a GPU or any local infrastructure?
No. WaveSpeed hosts both models on optimized infrastructure. Send a request, get a result. Nothing to provision, nothing to maintain.
How do I choose between 1080p and 4K output?
Match your delivery spec. 1080p at $0.025 per 5 seconds covers most web and social publishing. Go to 4K for large-format display or archival output you will reuse. One clean pass from source always beats re-upscaling later.