UNIT) = $0.01
inference:image:generatefal-ai/flux-pro/kontext/max/multi10 units · $0.10fal-ai/flux-2-pro4 units · $0.04fal-ai/nano-banana5 units · $0.05inference:3d:image-to-meshfal-ai/trellis3 units · $0.03fal-ai/hunyuan3d/v219 units · $0.19inference:3d:text-to-meshfal-ai/hunyuan-3d/v3.1/pro/text-to-3d44 units · $0.44inference:video:generatefal-ai/kling-video/v2.1/pro/image-to-video12 units · $0.12fal-ai/bytedance/seedance/v1/pro/image-to-video14 units · $0.14fal-ai/minimax/hailuo-02/standard/image-to-video6 units · $0.06
Open
Best-effort, no privacy claim
At base ratesthe default
- Any hardware, anonymous donor machines
- Runs whole on one machine — its operator can see your prompt and output
- Great for public or synthetic data, batch & benchmarking
- Not private — no confidentiality control
Trusted
Operator vetting + contract
Above basevetted capacity
- Input & output slices go to vetted operators
- Cheap anonymous capacity fills the middle slices
- Vetting is the bar — a vetted operator on a consumer card qualifies
- Built for real customer workloads
Popular
Private
Cryptographic — you verify, not trust
Premiumconfidential
- Confidential VM on a confidential-computing GPU
- System memory, GPU memory & the link between them encrypted
- The operator — us included — cannot read your work
- GPU-signed attestation, checked before any weight ships