Video generation, then restoration
Pair WAN video generation with a SeedVR2 upscale
WAN generates the clip; SeedVR2 restores and enlarges it. The two steps have different failure modes, and running them as one blurred job hides both. Generate the clip with WAN, keep the segment with clean motion, then upscale that keeper with SeedVR2 using 4n+1 batches. Test a short excerpt before committing the full clip.
This is a SeedVR2.net pipeline explainer. WAN is an upstream video-generation model family and SeedVR2 is a restoration model; this page describes how to chain them, not a benchmark of either. The hosted tool is a separate SeedVR2.net service.
Upscale the keeper clip without local batches
Once you have a generated keeper clip, upload a short excerpt here and read the credit preflight. The hosted video lane handles duration, model, and temporal cleanup without local batch tuning.
First eligible 5-second 720p Standard run: free
Verify your email to use your first video run without credits or a card. A failed run does not use the trial; longer clips and other settings show their exact credit cost before submission.
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 video pipeline, by the numbers
These four points capture the WAN to SeedVR2 video loop. They are workflow guidance, not a benchmark of either model.
- Step one
- WAN video generation
- Step two
- SeedVR2 upscale
- Batch pattern
- 4n+1 (5, 9, 13, 17)
- Test first
- A 5-second excerpt
Generate the clip and pick the segment with clean motion.
Restore and enlarge the keeper, not every take.
Follow the 4n+1 pattern for temporal consistency.
Test the hardest excerpt before the full clip.
Generation and restoration fail differently
WAN's failure mode is motion: broken limbs, flicker, and objects that morph between frames. SeedVR2's failure mode is restoration: over-sharpening, invented texture, and identity drift. Running them as one step hides which one produced a bad frame.
Keep the two steps separate so a bad result has a clear owner. Generate the clip, judge the motion, then upscale the segment that already moves cleanly.
Keep the segment with clean motion
Upscaling does not fix broken motion. If the generated clip already flickers or morphs, SeedVR2 will make the artifacts sharper, not smoother. Pick the segment where motion is coherent before upscaling.
A single clean segment beats a batch of enlarged bad takes. Route the source to the right step first, then spend credits only on the keeper.
- Generate the clip first.
- Keep the segment with clean motion.
- Upscale only the keeper.
4n+1 batches for temporal consistency
For local video upscaling, batch size controls how many frames are processed together, which drives temporal consistency. The SeedVR2 workflow expects video batches to follow the 4n+1 pattern such as 5, 9, 13, or 17 frames.
Do not shrink the batch below the useful range to chase a longer clip. A tiny batch can reduce flicker protection and leave frame-to-frame drift, which is exactly the artifact a video pipeline exists to avoid.
Test a short excerpt before the full clip
Duration is cost in a video pipeline. Extract the five seconds that contain the hardest motion and upscale that excerpt first. If the excerpt passes at normal speed, the full clip is a reasonable next step; if it fails, you saved the cost of a long job.
Review several timestamps, not one flattering paused frame. Motion quality is judged by playback, and a single sharp frame can hide flicker that appears the moment the clip moves.
When to use this pipeline
Use the WAN to SeedVR2 loop when you generate short AI clips and need a cleaner, larger final export. It is the generation-to-delivery path for creators who already have a base clip.
If you are starting from a real compressed video rather than a generated clip, skip the generation step and go straight to video upscaling.
Pipeline guidance limits
This describes how to chain WAN video generation with SeedVR2 upscaling, not a benchmark of either model. Model behavior changes; validate the loop on your own source and hardware.
- WAN and SeedVR2 versions change independently.
- Upscaling cannot fix broken motion the generation produced.
- Tiny video batches reduce temporal stability.
WAN to SeedVR2 questions
Should I generate with WAN and upscale with SeedVR2?
Yes, in that order. Generate the clip with WAN, keep the segment with clean motion, then upscale that keeper with SeedVR2. Keeping the steps separate makes a bad result easier to diagnose.
Will SeedVR2 fix broken motion from WAN?
No. If the generated clip already flickers or morphs, SeedVR2 will sharpen the artifacts, not smooth them. Pick a segment with coherent motion before upscaling.
What batch size should I use for the upscale?
Follow the 4n+1 pattern such as 5, 9, 13, or 17 frames. Do not shrink the batch below the useful range to fit a longer clip, because tiny batches reduce flicker protection.
Should I upscale the full clip at once?
No. Extract the five seconds with the hardest motion and test that excerpt first. Duration is cost, and a passing excerpt is the cheapest proof the full clip is worth running.