AI video upscaling vs ffmpeg scaling: when a model beats a filter

AI video upscaling reconstructs detail with a trained model, while ffmpeg scaling resamples existing pixels with a filter (for example bicubic or lanczos). ffmpeg is faster and free for straightforward resolution changes, but it cannot add detail that was never captured. A model such as Seedance native 4K, priced at about $4.90 per 5-second clip on relaydance.com/models, generates or reconstructs frames when a resample cannot recover missing information.

When ffmpeg scaling is enough

Use ffmpeg when you only need to change resolution or fit a display, not recover lost detail. A filter such as scale=1920:1080 with lanczos interpolates between known pixels, runs locally, and adds no per-second cost. This suits format conversion, container repackaging, and downscaling from a higher-resolution source. ffmpeg cannot invent texture that the original recording never held, so upscaling low-resolution footage with a filter tends to produce soft or blocky output. For predictable, low-cost resizing where the source already contains the needed detail, a filter is the practical option and avoids API billing entirely.

When a model beats a filter

A model beats a filter when the source lacks detail and you need reconstruction rather than resampling. Seedance 2.0 renders at 1080p for about $0.470 per second and at 720p for about $0.190 per second on relaydance.com/models, while Seedance Fast is about $0.152 per second. Seedance native 4K costs about $4.90 per 5-second clip. These rates reflect generation work that a filter does not perform. Because the API is pay-as-you-go, cost scales with duration and resolution. According to the official relaydance.com docs, 「Failed or errored requests are never billed」, so a rejected job adds no charge to your reconstruction budget.

Cost and capability comparison

The comparison below contrasts a local filter with model-based options and their published rates.

How to submit a model job

Submit a model-based reconstruction job through the OpenAI-compatible API in a few steps. According to the official relaydance.com/docs, 「Set base_url to https://relaydance.com/v1 and keep your OpenAI SDK」.

1. Create an API key at https://relaydance.com/console and set Authorization: Bearer YOUR_API_KEY.
2. Point base_url to https://relaydance.com/v1.
3. POST /v1/video/generations with model, prompt, seconds, and metadata (ratio, resolution, generate_audio, callback_url).
4. Add reference media in metadata.content[] and cite them in the prompt as @image1 to @imageN (up to 9 images, 3 videos, 3 audio tracks; clips up to 15 seconds).
5. Poll GET /v1/video/generations/{task_id} until status is succeeded or failed, or set metadata.callback_url for webhook delivery.

See relaydance.com/docs for the full request reference.

FAQ

Does ffmpeg add detail to low-resolution video? No. ffmpeg resamples existing pixels with a filter and cannot recover detail the source never captured; a trained model performs that reconstruction.

Do I pay for jobs that fail? No. According to the official relaydance.com docs, 「Failed or errored requests are never billed」, and billing is pay-as-you-go per generated video.

Can I keep my existing OpenAI SDK? Yes. Set base_url to https://relaydance.com/v1 and use Authorization: Bearer YOUR_API_KEY; the API is OpenAI-compatible.