VEnhancer
One diffusion framework that does spatial upscaling, temporal upscaling, and artifact cleanup together — built to polish AI-generated video.
About
VEnhancer (arXiv 2407.07667) is a generative space-time enhancement framework from Jingwen He's team. Built on a pretrained video diffusion model with a trainable video ControlNet injected as conditioning on low-frame-rate, low-resolution inputs, it does three jobs in one pass: spatial super-resolution at arbitrary scales (1x to 8x, with 3x-4x recommended), temporal super-resolution to a target frame rate, and refinement that strips spatial artifacts and temporal flicker — the exact defects that plague AI-generated video.
Space-time data augmentation and video-aware conditioning keep the end-to-end training stable. The authors report it beating existing video and space-time super-resolution methods on AI-generated clips, and claim it lifted VideoCrafter-2 to the top of VBench (their numbers). The official repo ships inference code, a Gradio demo, and checkpoints on Hugging Face. Real limitations, stated by the authors themselves: diffusion inference is slower than one-step methods, and long videos over about 10 seconds can drift.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported (arXiv 2407.07667): surpasses existing VSR and space-time SR methods on AI-generated video benchmarks; VideoCrafter-2 + VEnhancer claimed top-1 on VBench
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: paper arXiv 2407.07667 (He et al.); official repo ueoo/venhancer confirms inference code, Gradio demo, and HF checkpoints. Strengths: unified space-time enhancement, open weights and code, honest author-stated limitations. Docked for: (1) benchmark claims are author-reported; (2) checkpoint license terms not stated; (3) diffusion inference is slow and long-video consistency is an acknowledged weakness. License: research code; checkpoint license terms unverified.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · arXiv 2407.07667 (He et al.); video diffusion + trainable video ControlNet
- · Unified 1-8x spatial SR, temporal SR, and artifact/flicker refinement
- · Official repo ueoo/venhancer: inference code, Gradio demo, HF checkpoints (jwhejwhe/VEnhancer)
- · Author-reported gains; VideoCrafter-2 VBench top-1 claim is provider-reported; slow inference; long videos acknowledged weak