The distributed
canvas of GPU compute.
CANAS.NET weaves millions of idle consumer GPUs into one mesh that renders generative diffusion — images and video — and settles it per frame on-chain. Requesters pay only for what they generate. Providers earn from silicon that would otherwise sit dark.
Submit a job. Watch the mesh render it.
This is the real thing: your prompt is priced in USDC, escrowed, routed to an actual provider in the live network, and rendered. It even shows up in the public job stream below.
Six steps from prompt
to settled frame.
CANAS borrows HTTP 402: an unpaid request gets a payment challenge, the consumer attaches proof, and the network does the rest — match, render, verify, settle.
Lock escrow
A job — prompt, model, sampler, seed — is submitted and payment is locked in a Solana escrow program. No accounts, no subscriptions.
Match a GPU
The scheduler routes by GPU model, VRAM, latency, price and reputation — to one provider, or several for premium jobs.
Render
The provider pulls a pinned container and content-addressed weights (Arweave/IPFS + CDN) and runs the diffusion inference.
Attest
The result is returned with an attestation: job id, parameter hash, model hash, a perceptual hash of the output, and a timestamp.
Verify
A sampled fraction is re-run by an independent verifier and compared by perceptual closeness. Canary jobs catch fabrication.
Settle
On acceptance the escrow releases — ~90% to the provider, ~5% to verifiers, ~5% to the protocol. Paid straight to wallet.
The network, rendering right now.
Every row below is a real diffusion job moving through escrow, scheduling, render and settlement — streamed straight from the mesh.
Live job stream
a lone astronaut drifting above a ringed planet, unreal engine render, bioluminescent ambiance, ray traced reflections, intricate detail
a witch brewing potions in a candle-lit cottage, isometric diorama, bioluminescent ambiance, depth of field, sharp focus, 8k
a clockwork owl perched on a gear-driven tree, pixel art, depth of field, volumetric fog
a lone astronaut drifting above a ringed planet, comic book ink, overcast diffuse light, depth of field, bokeh, wide angle
a tiny robot tending a rooftop garden, cinematic photography, golden hour, award-winning
a haunted lighthouse in a violet storm, vaporwave, moody chiaroscuro, wide angle
a street photographer in 1970s Tokyo, synthwave poster, soft rim lighting, depth of field, shot on Hasselblad
a cozy bookshop on a snowy evening, oil painting, moody chiaroscuro, 35mm
a desert nomad beneath two moons, analog film photo, bioluminescent ambiance, shot on Hasselblad
a brutalist temple carved into a glacier, hyperrealism, soft rim lighting, intricate detail, volumetric fog
a jellyfish made of stained glass, cinematic photography, bioluminescent ambiance, sharp focus, bokeh
a derelict space station orbiting a gas giant, unreal engine render, intricate detail, octane render, wide angle
a floating city above an endless ocean, vaporwave, volumetric god rays, 8k, award-winning, ultra-detailed textures
a phoenix rising from molten obsidian, low-poly 3D, iridescent backlight, film grain, sharp focus
a derelict space station orbiting a gas giant, matte painting, candlelit warmth, 35mm
Provider map
24 regionsTop providers
by earningsWe don't pretend to prove a GPU.
We make lying unprofitable.
Bit-exact proof of diffusion across heterogeneous hardware is physically impossible. So CANAS trades cryptographic certainty for an economic & statistical one — reproducibility by closeness, canaries, redundancy, reputation and slashing.
Perceptual reproducibility
LPIPS · SSIM · pHashA verifier with the same pinned container re-runs the job and compares outputs by LPIPS / SSIM / pHash within tolerance — never bit-for-bit, because floating point diverges across hardware.
Canary jobs
known seed · known outputThe scheduler blends in reference prompts with a known seed and reference output. Any node fabricating or substituting results fails the canary and gets caught.
Redundant execution
N-of-M consensusPremium and high-value jobs run on several providers at once. Consensus is the perceptual cluster — outliers are rejected before settlement.
Optimistic + sampling
sampled · slashableMost jobs settle without re-checks; a random fraction is re-verified, and disputes resolve by redundant re-execution. Dishonesty is met with bond slashing.
honest limit — this is an economic and statistical guarantee, not a cryptographic proof of computation. For subjective visual output it's a fair trade; for safety-critical compute it isn't the right tool, and CANAS doesn't claim to be.
Your idle GPU is a render farm.
Most gaming GPUs sit dark for 18+ hours a day. Point yours at CANAS and it renders diffusion jobs for people who'd otherwise pay a centralized cloud — and you keep ~90% of every settlement.
Earnings estimate
indicativePaid in USDC, settled to your wallet each epoch · ~90% provider / 5% verifier / 5% protocol · estimates scale with real network demand.
Install the worker
One Docker container — diffusers / ComfyUI, CUDA, pinned versions. A light agent opens a reverse tunnel.
Stake a bond
Lock $CNSNET as a provider bond. It backs honest output and is slashable for fabrication.
Go online & earn
The scheduler routes jobs to you by GPU, VRAM and reputation. USDC settles to your wallet each epoch.
$CNSNET — a utility, not a yield machine.
Users pay in USDC for price stability. $CNSNET secures the network: it bonds providers, stakes verifiers, governs parameters, and unlocks priority. Value tracks job flow — not emissions.
Utility
- Provider bond — slashed for fabricated or substituted output
- Verifier stake — collateral for honest re-checks
- Governance — network params, fee rates, model set
- Priority routing & reduced fees for holders
Token allocation
Settlement split per job
214 providers bonded · paying protocol fees in $CNSNET costs less than in stablecoin.
From test mesh to planetary canvas.
Test mesh
- Permissioned provider set
- Baseline image generation
- Escrow on Solana devnet
Mainnet-beta
- Open provider onboarding & reputation
- Canary + sampled verification
- USDC settlement, per-frame pricing
Scheduler decentralization
- Open verifier set
- $CNSNET governance
- Video diffusion workflows
Maturity
- Redundant execution for premium jobs
- Expanded model catalog
- SLA tiers
Paint with the world's
idle GPUs.
Render diffusion for a fraction of cloud prices, or turn your gaming rig into a cash-flowing render node. Two sides, one canvas.