Train and run your models
Training, fine-tuning and batch inference with your own Python/PyTorch project. Transcription is also available.
Future direction
Room to grow.
Proof before promises.
Where ComputeMint is heading. These are plans and ideas, not a launch schedule or an existing marketplace.
Images, 3D and narrated video
Upscaling → image generation and editing → image → 3D → notes → narrated video → video generation. All five are planned, not available yet. No launch dates are set.
Blender rendering
Render a Blender scene to frames on the GPU. File packaging, add-ons and resource limits all need testing first, so this is an idea, not a plan.
Independent providers
A possible multi-provider platform, with supported worker software, clear provider trust limits and evidence-backed machine selection. It does not exist today.
What needs to hold up as we grow
The platform should outlast a worker outage.
Move accounts, job state and retained results off the GPU laptop before relying on independent providers. A disconnected machine should not take the whole customer experience offline.
A cheaper hour does not always mean a cheaper job.
Future machine choices should explain supported workloads, measured performance, quote terms and uncertainty. Accepted prices need to stay fixed under their quoted rules.
Provider trust needs a clear explanation.
A provider controls their host and can potentially inspect customer data. Container isolation does not hide plaintext from that host’s administrator. Future choices need explicit data-handling and trust limits before upload.
Reliability needs evidence.
Show observation periods, sample sizes and failure attribution before publishing scores. Decorative badges or invented success percentages do not establish reliability.
Customer credits and provider earnings are different.
Provider accounting, reconciliation, disputes and payout readiness need separate work. The current test-mode credit flow does not establish a marketplace payout system.