Vidaio
SN85AI-powered video processing that can enhance, analyse, and transform video at scale
The subnet that turns low-resolution video into 4K and crunches large files down to a fraction of their size, with miners competing on objective video-quality metrics rather than subjective taste.
// Decentralized upscaling and compression.
Vidaio is a Bittensor subnet for AI-driven video processing. Miners run upscaling and compression models that take low-resolution or large-file videos and return higher-resolution or smaller-file outputs. Validators benchmark each miner using established perceptual quality metrics, then weight the network accordingly.
The simple version: Like the upscaling feature in NVIDIA's RTX video tools, but distributed across a network of miners competing to do it best.
Centralized equivalent: Think Topaz Video AI for upscaling and Mux or Bitmovin for compression, but as a single open network anyone can plug into.
How it works:
- Miners receive video chunks, either from validator benchmarks or real user uploads, and run their models to upscale low-resolution clips back to high resolution, or compress high-quality clips while preserving visual fidelity.
- Validators score outputs using VMAF and PieAPP, two industry-standard perceptual quality metrics, and rank miners on accuracy for upscaling and on quality-to-size ratio for compression.
- The problem it solves: Video enhancement is centralized and expensive. Desktop tools like Topaz Video AI charge hundreds of dollars per seat, and cloud transcoding services bill by the minute. Smaller creators and platforms get priced out of high-quality video pipelines.
- The opportunity: Every platform that touches video, from streaming and social to archives and restoration, needs cheap, scalable processing. The workload is uniform enough to parallelize across many miners.
- The Bittensor advantage: VMAF and PieAPP are well established in the video industry, so the incentive mechanism rests on measurable quality output rather than subjective judgment. That makes it cleaner to score than most generative AI subnets.
- Traction signals: 100 registered miners on-chain, seven contributors merging pull requests in the last month, and a six-phase roadmap that progresses from current upscaling and compression toward on-demand streaming and a public REST API.
Category: Image/Video/Audio Generation | Centralized Competitor: Topaz Video AI, Mux, Bitmovin, AWS Elastic Transcoder
Vidaio occupies a specific lane: pure video post-processing. Most subnets in the generative AI category create content from scratch, whether text, image, or 3D. Vidaio improves content that already exists. That distinction matters because the work is more measurable and the customer demand is more concrete.
Mechanism:
The subnet handles two task types. For upscaling, validators take a high-resolution source video, downscale it, and send the low-resolution version to miners. Miners run their models to upscale it back to the original resolution, and validators measure how closely the result matches the original using VMAF and PieAPP. For compression, validators send high-quality source video to miners, who must produce a smaller file while preserving as much perceptual quality as possible. The quality-to-size ratio determines the score.
Organic queries follow the same pipeline but use real user uploads. Videos are chunked, queued, distributed to miners, processed, and reassembled. This is the path Vidaio needs to scale to make the subnet self-sustaining beyond emission rewards.
Development cadence is steady. The GitHub repository at vidaio-subnet/vidaio-subnet has 734 commits across 20 contributors, with the most recent push on 2026-05-28, two days ago. Over the past month, seven contributors merged work covering organic-query scoring, a validator dev mode, authentication for organic endpoints, polling-based organics, a Docker Compose miner setup, and wandb logging fixes. Arpan Tripathi authored most of the recent merges.
Market position is more challenging. At 0.00875 TAO, the alpha token has fallen 34.6% over thirty days and 17.6% over ninety days, leaving a market cap of 37,973 TAO and pool depth of 15,999 TAO. Root proportion sits at 0.18, meaning most of the pool reflects organic staking rather than protocol subsidy. Net seven-day flow is negative 221 TAO. Modest outflows, but enough under Taoflow's EMA-smoothed model to leave the subnet's current emission share at 0%. The subnet still runs, with validators scoring and miners working, but no new TAO is being injected into the pool reserve until net flows turn positive again.
The roadmap is ambitious. Phase three introduces transcoding optimization, phase four enables on-demand decentralized streaming with integrated storage, phase five tackles live streaming with real-time AI processing, and phase six exposes a RESTful API for external integration. The current state is mid-phase two: refining the core upscaling and compression mechanisms with adaptive-bitrate work flagged next.
- Zero emission share currently: Under Taoflow, sustained negative net flows have pulled Vidaio's emission share to 0%. Miners and validators are working without fresh TAO inflow until staking demand reverses. That is a real squeeze on participant economics, even with active development.
- Concentrated contributor base: Arpan Tripathi authored most recent commits, and historical contributions are heavily weighted to one earlier lead. Bus factor remains a concern for a project of this scope.
- Competition for video AI workloads: Other subnets and centralized services target the same demand. Vidaio's quality-to-cost edge needs to show up in real organic traffic before it can defend a position.
- Thin liquidity: With 16,000 TAO in the pool and 841 TAO of twenty-four-hour volume, the market is thin. Larger trades will move price meaningfully.
Into the next one.
