One video in. The same video out, at the resolution you meant to shoot it in.
SeedVR2
Detail, not sharpening
A sharpening filter pushes contrast at the edges it can already see. SeedVR2 rebuilds what the pixels were standing in for: pores, knit stitches, the chalk on a menu board. Pick a spot and look at it at 300%.
Whatever it came from
Low resolution has a few usual causes. SeedVR2 was trained on all of them, so the same pass handles a generated draft, a phone clip and a tape transfer.
Generated at 480p
Draft fast and cheap in Seedance or Hailuo, then finish the pick. The typical generation smear comes off with the softness.
Shot on a phone
Compressed, handheld, indoor light. Blocks and mush go; the take stays exactly the take.
Pulled from the archive
Camcorder and DV transfers: soft, noisy, colors bleeding. You get a clean master that will take a grade.
Draft low. Deliver high.
Generation cost climbs with resolution. Make ten versions at 480p, choose the one that works, and run only that one through SeedVR2.
- 01Generate drafts at 480pAny model on Plot Party. Iterate on the prompt while it is cheap.
- 02Pick the takeOne clip earns the finish. The rest cost you a few points each.
- 03Upscale onceSeedVR2 returns the same clip at 1080p. Same timing, same motion, same cut.
480P
480P
480P
480P
480P
480P480PEvery size on the ladder
Output frames drawn at true relative scale. Your 480p draft is the small box in the corner; the 1080p master has five times the pixels.
- IN
- 480p · 854×480
- OUT
- 1080p · 1920×1080
- PIXELS
- ×5.1
Upscaler vs. SeedVR2
What a conventional upscaler does to a frame, and what SeedVR2 does instead.
- Method
- UPSCALERInterpolates between the pixels it has, then sharpens the resultSEEDVR2Generative restoration: predicts the detail the source lost, guided by every frame around it
- Fine detail
- UPSCALERLarger and softerSEEDVR2Rebuilt: skin, lettering, fabric, foliage
- Compression artifacts
- UPSCALEREnlarged along with everything elseSEEDVR2Removed
- Frame to frame
- UPSCALEREach frame filtered on its own; detail can shimmerSEEDVR2Consistent across the clip, no flicker
- AI-generated footage
- UPSCALERThe same generation smear, biggerSEEDVR2Trained on it; the smear comes off with the softness
- What you do
- UPSCALERPick a filter, tune a strengthSEEDVR2Upload the clip. One pass.
| ASPECT | CONVENTIONAL UPSCALER | SEEDVR2 |
|---|---|---|
| Method | Interpolates between the pixels it has, then sharpens the result | Generative restoration: predicts the detail the source lost, guided by every frame around it |
| Fine detail | Larger and softer | Rebuilt: skin, lettering, fabric, foliage |
| Compression artifacts | Enlarged along with everything else | Removed |
| Frame to frame | Each frame filtered on its own; detail can shimmer | Consistent across the clip, no flicker |
| AI-generated footage | The same generation smear, bigger | Trained on it; the smear comes off with the softness |
| What you do | Pick a filter, tune a strength | Upload the clip. One pass. |
Spec sheet
SeedVR2 is ByteDance Seed's video restoration model. On Plot Party it runs in the browser generator and through the API and MCP, so the same clip you finish here can be finished from Claude, ChatGPT or your own pipeline.
- Model
- SeedVR2 by ByteDance Seed · model id seedvr2
- Task
- Video to video. One source clip in, the same clip out at higher resolution
- Input
- AI-generated or camera footage at 480p or 720p, mp4 or mov
- Output
- 1080p. Same length, same frame rate, same framing as the source
- What changes
- Resolution, fine detail, compression noise and generation smear
- What stays
- Timing, motion, composition, color and the cut. Nothing is regenerated
- Prompt
- None. The model reads only the video
- Access
- Plot Party generator, REST API and MCP server
Questions
What is SeedVR2?
SeedVR2 is a video restoration model from ByteDance Seed. You give it one video and it returns the same video at a higher resolution, with fine detail restored and compression noise removed, while keeping the motion consistent from frame to frame. On Plot Party you can run it in the browser generator or call it through the API and MCP with the model id seedvr2.
Is this the same as an upscaler or a sharpening filter?
No. An upscaler interpolates between the pixels it already has and then sharpens the edges, so the result is the same picture, larger and a little crisper. SeedVR2 is generative: it predicts the detail the source lost, using the surrounding frames as evidence, so skin, lettering and fabric come back as texture rather than as contrast.
Does it work on AI-generated video?
Yes, and that is the main use on Plot Party. Generate drafts at 480p with any model, pick the one that works, and run that one through SeedVR2 to get a 1080p master. The typical softness and smear of generated footage is one of the things it was trained to remove.
Will it change what happens in my video?
No. SeedVR2 does not regenerate the clip. Timing, motion, framing, color and the cut are preserved; only resolution and detail change. The output has the same length and frame rate as the source.
What input and output resolutions are supported?
Send 480p or 720p footage and get 1080p back. The output keeps the aspect ratio of the source.
Does it keep the audio?
The audio track from your source video is passed through unchanged.
How is SeedVR2 billed on Plot Party?
You pay in Plot Points per second of source video, so a 10-second clip costs twice a 5-second one. Current rates and plans are on the pricing page.
Is SeedVR2 available through an API?
Yes. The Plot Party API and MCP server expose SeedVR2 as seedvr2. Upload the source video first, then pass its URL as the input. The API docs list every parameter.
Look closer.
Send the clip you already have. Get back the one you meant to make.