The hosted SeedVR2 upscaler
SeedVR2 upscaler for videos and images, online
Start with the file you have. For video, open the panel below with SeedVR2 at 1080p, add a short clip and check credits before rendering. For a photo or illustration, use the image upscaler instead. Both run through a hosted service without a local GPU or ComfyUI installation; your source is uploaded for processing.
A documented SeedVR2 image example
This original vector illustration was rendered to a 768 × 768 input and processed by the SeedVR2 image model to 2048 × 2048 on July 20, 2026. It shows one image result, not a video benchmark.
We authored the illustration; this is not a customer photograph, a filmed scene or evidence of video motion quality. The source record below lists the input and output hashes, dimensions and model.
SeedVR2.net is an independent paid service that uses hosted model providers; it is not ByteDance’s official research site. This page opens the SeedVR2 video model at 1080p, while the separate image tool handles photos. Our recorded illustration below demonstrates image processing only, not video quality or a customer result.
Run the SeedVR2 upscaler in the browser
This panel opens with SeedVR2 and 1080p selected. Add an MP4, MOV or WebM file up to 500 MB, or a direct public HTTPS video URL, then check the duration, model and credit quote. Video processing uses credits; this page does not promise a free 2K video trial. Nothing is rendered until you submit.
Browser upload limit: 500 MB. For larger files, paste a publicly accessible HTTPS URL. Each job supports videos up to 10 minutes.
Quick scene presets
Presets pick a model for a common scenario. You can still switch models manually below.
Upscaled preview
Compare the enhanced output before exporting the final file.
Upload a clip, paste a direct video URL, or load the sample. No example video is downloaded until you ask for it.
Tip: keep your original FPS and avoid pre-scaling. The upscaler handles cadence better than most NLE plugins.
Test before processing the full video
Upscaling cannot recreate detail removed by heavy messaging compression. Test the hardest 5–10-second segment first; a higher-priced model is not automatically a better fit.
The upscaler at a glance
Five facts decide whether the browser upscaler fits your job: the model behind it, what you can submit, which resolutions exist, what it costs, and what you download at the end.
- Model
- SeedVR2 (ByteDance)
- Input
- 500 MB or a public URL
- Resolutions
- 720p / 1080p / 2K / 4K
- Cost
- Credit preflight first
- Output
- Downloaded video file
This page selects the hosted SeedVR2 video model. Standard, FlashVSR, Pro and Ultimate are separate model choices.
Browser upload up to 500 MB, or a directly accessible public HTTPS video URL, up to ten minutes.
All five video models offer these targets. Select the model and resolution separately, then read the quote.
The tool shows required credits before submission. Video processing is paid; any account-specific offer must be confirmed in the tool.
You download the upscaled file when the job finishes; nothing is installed locally.
What the SeedVR2 upscaler actually is
SeedVR2 is ByteDance's restoration model for low-resolution, compressed, or damaged video and images. The upscaler on this site is that model exposed as a hosted web tool. The key difference from running the research code yourself is that the inference, GPU, and model weights live on a provider, so you only need a browser, an upload path, and enough space to keep the result.
The model restores rather than simply resizes. A naive upscale stretches existing pixels; SeedVR2 reconstructs plausible detail from what is left in the source. That is why a low-res face or small text can come back sharper, and why already-erased pixels cannot be invented truthfully. Restoration improves damaged signal; it does not recover information that was never recorded.
- SeedVR2 is a restoration model, not a pure resize filter.
- Hosted inference means no GPU, no model files, and no local graph.
- Restored detail is plausible, not a factual recovery of lost pixels.
How to upscale a clip, step by step
Open the hosted upscaler and add a source: either upload a browser file up to 500 MB or paste a directly accessible public HTTPS video URL. The tool reads duration and shows which resolutions are available. Sources longer than ten minutes are outside the supported range, so trim a long master before submitting.
Choose the model and target separately: SeedVR2, Standard and Pro can each target 4K, but use different credit rates. This page starts at SeedVR2 1080p. Read the quote for your verified duration before submitting; opening the tool or reading a quote does not render a video.
- Add a file or a public HTTPS URL, then read the duration.
- Choose 1080p for general delivery; reserve 4K for sources with real detail.
- Confirm the credit preflight before submitting a paid job.
Picking the right resolution and model path
Standard, FlashVSR, SeedVR2, Pro and Ultimate each expose 720p, 1080p, 2K and 4K. The largest target is not automatically the best result. Match the output to the screen or delivery requirement, and compare a short difficult excerpt before processing a longer video.
Think in terms of deliverable, not fantasy. A 1080p deliverable for a phone or a social feed is a defensible target; a 4K master for a cinema finish only matters if the source, lens, and pipeline support it. When in doubt, run one short 1080p sample first and judge the result before committing the full clip.
What the upscaler can and cannot fix
Low resolution, noise and compression are reasons to try restoration, not promises of a clean result. Check skin, fine lettering, hair, line art and motion on your own clip. A sharper edge can still be a wrong edge, and smoothing can remove detail you wanted to keep.
It cannot recover a face that was a four-pixel blob, read text that was never legible, or restore motion detail that was blurred out at capture. It can also invent detail that looks confident but is not in the source. For important material, review faces, text, and motion frame by frame before you accept the file.
- Candidate sources: low resolution, noise or compression; test each source.
- Hard cases: tiny faces, illegible text, capture-time motion blur.
- Treat invented detail as a reason to review, not a reason to skip review.
Reviewing an upscaled clip before you trust it
Download the finished file and review it at delivery size and normal speed first, then scrub frame by frame. Check that identities are unchanged, small text is accurate, hair and fabric keep texture instead of smoothing into a single tone, and motion stays consistent from frame to frame. A single marketing still is not enough to judge a video.
Keep the source alongside the output. If a crop fails, change one thing at a time, such as the resolution or the source trim, and re-run only the failing section. Rejecting and re-running the whole clip repeatedly is the expensive way to hunt for a setting that was never the problem.
SeedVR2 upscaler limits
Upload limits, resolutions, credit amounts, and hosted behavior change. This guide reflects the hosted tool at review time; always confirm the live preflight and file limits before you submit a paid job.
- Browser uploads are capped around 500 MB; larger sources need a public direct URL.
- Sources longer than about ten minutes fall outside the supported range.
- Restoration cannot recover detail that the source never captured.
- The hosted path depends on network access and provider availability.
Sources and sample provenance
Checked October 24, 2026. Research documentation describes the model; the live tool defines this hosted service. A recorded sample is not a quality guarantee.
- ByteDance SeedVR research repository
Original research code and documentation for SeedVR and SeedVR2. SeedVR2.net is a separate hosted service, not this research project.
- Official SeedVR2-3B model card
The upstream model card documents the SeedVR2 name and Apache-2.0 license; it does not promise identical results from every hosted integration.
- Read the sample source record (JSON)
Original illustration, actual model output, file hashes and measured dimensions. This is image evidence only.
SeedVR2 upscaler questions
Do I need a GPU to use the SeedVR2 upscaler?
No. The hosted upscaler runs inference on a provider, so your device only needs a current browser and a reliable connection. A local GPU is only required if you self-host the research code or run the ComfyUI template.
Which resolution should I choose, 1080p or 4K?
Choose 1080p for a social or delivery master and reserve 4K for sources that genuinely have recoverable detail. Upscaling a low-detail source to 4K spends more credits without adding information, so run a short 1080p sample first.
How long can my source clip be?
The hosted tool supports sources up to about ten minutes. Trim longer masters before submission, and note that the credit preflight scales with model variant, billed duration, and resolution.
What do I download at the end, and is there an app to install?
You download the finished upscaled video file. There is no desktop app to install; the upscaler runs in the browser. The only separate local path is the open-source model for ComfyUI, which is a different workflow.