Frequently Asked Questions

MachGen is an AI inference platform purpose-built to run next-generation generative and perception models with ultra-low latency and dramatically lower cost. It is optimized for image and video generation, visual perception, and emerging world models - helping developers and enterprises deploy these workloads at production scale.

MachGen serves a broad range of users: developers and AI agents who want simple API access to top models, prosumers and hobbyists who need high-quality video generation on demand, and enterprises that require a fully optimized inference stack for their own proprietary models. If speed, quality, and cost matter to you, MachGen is built for you.

Getting started takes minutes. Create an account, purchase credits, and start calling our API immediately, with no infrastructure setup required. For enterprise customers looking to deploy the MachGen Platform, reach out to our team via the Contact page and we'll walk you through the onboarding process.

Most platforms are wrappers around the same model weights. MachGen goes deeper: our engineering team optimizes the entire inference stack, from kernel-level GPU utilization to batching and scheduling, to deliver inference speeds and costs that off-the-shelf deployments simply can't match. We then offer that optimized stack to you, either via our serverless API or deployed directly in your environment.

Not at all. MachGen Serverless is designed for simplicity: pick a model, send a prompt, receive a video. Our REST API and documentation make integration straightforward for any developer. For enterprises leveraging the full MachGen Platform, our technical team works alongside yours to handle the optimization and deployment complexity.

MachGen Serverless gives you instant, pay-as-you-go access to our full suite of optimized video generation models via a simple API. You purchase credits and spend them across any model. No commitment, no infrastructure to manage. It's ideal for developers, agents, and anyone who wants world-class video generation without the operational overhead. If you're interested in an enterprise contract (additional discounts, support), please contact us.

MachGen Platform is our enterprise offering. We work closely with your team to optimize the inference stack for your specific models and use case, then deploy it in the configuration that suits your business: MachGen Cloud (fully managed by us) or Customer-hosted (deployed inside your own infrastructure). The result is dramatically better latency and cost than running standard model deployments.

Credits are the currency of MachGen Serverless. Each API call deducts credits based on the model used, the output resolution, and other parameters like image resolution or duration of generated video. You'll see the price once you've selected the model and input/output parameters. Credits never lock you into a single model. Use them however and wherever you get the most value.

No. Credits you purchase do not expire. We believe you should be able to use what you pay for on your schedule, not ours.

Yes. For a limited time, verified new accounts receive a small credit allocation to explore the platform and test our models before entering their credit card information. For enterprise evaluations of MachGen Platform, contact our team and we'll set up a structured pilot.

We accept all major credit and debit cards for credit purchases. Enterprise customers can arrange invoiced billing.

MachGen supports a highly curated selection of open and commercial video generation models that offer superb price-performance ratio. We continuously evaluate new models for inclusion.

For open models like LTX, Flux.2-dev, Qwen 3-VL, GLM 4.5, etc., significantly faster than standard deployments. Our inference optimizations, including mega kernel fusions, efficient parallelism, dynamic caching, and other magic, deliver latency reductions of up to 80% compared to vanilla model deployments, depending on the model and hardware configuration. We're able to serve many of these at a much lower cost as well. Exact numbers vary by model; see our benchmarks in the documentation.

Supported resolutions and output lengths vary by model. Most models support 720p output, with select models reaching 1080p. Video durations typically range from 3 to 10 seconds per generation. Full per-model specs can be seen once you select a model to generate an output.

Yes. This is the core value proposition of MachGen Platform. Our team works with you to bring your proprietary or fine-tuned models onto our optimized inference stack. We handle the engineering complexity so you can focus on your product.

See "How fast is video generation on MachGen?" above.

MachGen Platform offers two deployment paths: MachGen Cloud, where we manage the full deployment on dedicated infrastructure so your team can focus on product growth; and Customer-hosted, where we deploy the optimized inference stack directly inside your own cloud environment or data center, giving you full control over data residency and security.

We take data privacy seriously. We never use inputs and outputs processed through MachGen Serverless to train or fine-tune any models. For enterprise customers on Customer-hosted deployments, your data never leaves your infrastructure. For MachGen Cloud deployments, ensure your data is not accessible to anyone outside of authorized personnel in your organization. For added peace of mind, we offer dedicated single-tenant environments on request.

We are actively working toward SOC 2 Type II certification. Enterprise customers with specific compliance requirements should reach out to discuss our current controls and roadmap.

Yes. MachGen Platform enterprise agreements include service level agreements covering uptime, response time, and support. The specific terms depend on your deployment configuration. Contact our team to discuss SLA options.

Reach out to our team via the Contact page. We'll schedule a call to understand your models, use case, and scale requirements, then scope an engagement. Typical onboarding for MachGen Platform takes one to three weeks depending on model complexity.

Yes. High-volume users and teams can access discounted credit bundles. Enterprise Platform customers receive custom pricing. Contact us to discuss volume pricing that fits your usage.