Flat-rate GPU vs. credits and per-second billing: what AI rendering really costs
There are three common ways to pay for AI rendering on someone else's GPUs. Each has its place. This guide shows where the costs come from, so you can do the math for your own workload.
- Credits, tokens or points: you buy a bundle of virtual currency and each job spends some of it.
- Per-second (or per-hour) GPU billing: you pay for the GPU time your jobs actually use.
- Flat rate: one monthly price covers unlimited jobs within the plan's features.
Credits: easy to buy, hard to read
With credits, every action has its own price tag. Higher resolution, video, longer clips, newer models and extras such as upscaling often cost more. A bundle of 1,000 credits does not tell you what one finished image costs, because the rate changes with every setting.
Per-second GPU billing: transparent, but variable
Paying per GPU-second is honest about what you buy: compute time. Your bill, though, follows everything that adds time:
- Cold starts and model loading: spinning up a GPU and loading a large model takes time before any rendering happens.
- Idle time: on some setups you pay while a GPU waits for the next job.
- Retries: a failed or rejected render uses the same GPU time as a good one.
- Heavy workloads: video and high-resolution workflows run longer, so each result costs more.
This model can be very efficient for steady, well-understood pipelines. For exploratory work it is hard to budget.
Where costs pile up on any metered plan
- Iteration: creative work means many drafts for each keeper.
- Learning and testing: trying new models, LoRAs and workflows is the work that gets metered most.
- Video: it needs far more compute than images.
- "Unlimited" with a slow lane: some services advertise unlimited plans, then throttle you and sell boosts. Read the terms.
How flat rate works on Graydient
Graydient charges one monthly price with no credits, tokens or boosts. Instead of metering every render, it lends each job a full, unpartitioned GPU for a set time: up to 3 minutes per job on Images and Video Plus, and up to 6 minutes on Video Max. You can batch up to 8 images at a time on Images and Video Plus, and up to 16 on Video Max.
Plan prices and limits are on the pricing page: Compare plans.
The trade-offs to know: GPU time per job and storage are capped by plan, and the personal API (Video Max) is rate-limited and for one user. Automated traffic is not allowed on personal plans. Bots, batch jobs, reselling or a live multi-user product need the Business plan. Prices are at business.graydient.ai. See the AI rendering API.
Do the math: a hypothetical example
Hypothetical example. The numbers below are invented for illustration and are not any real provider's prices. Plug in the real rates of the service you are comparing.
Say a metered service charges $0.002 per GPU-second (hypothetical) and one short video job takes 90 GPU-seconds, including model load:
- One render: 90 × $0.002 = $0.18
- You keep 1 clip in 4, so each keeper really costs 4 × $0.18 = $0.72
- For 60 keepers a month, that is 240 renders × $0.18 = $43.20
Then compare that with a flat plan that includes the video features you need. Ask yourself:
- What does one result cost? For credits, divide the bundle price by the number of credits, then multiply by the credits per job. For per-second billing, multiply the rate by the seconds per job, including overhead.
- How many attempts do you make per keeper? Count your retries, not only the results you keep.
- What is your monthly volume? Use your real habits and add a buffer for testing new models.
- Does a flat plan cover your workload? Check GPU time per job, resolution, storage and API access.
If your metered spend regularly goes over the flat price, flat rate wins, and every extra experiment costs nothing more.
When metered billing makes sense
- You render only occasionally.
- You need a GPU type, job length or scale beyond a flat plan's limits.
- You run a steady, tuned pipeline with few retries.
When flat rate makes sense
- You iterate a lot, test models often or make video.
- You want a predictable monthly bill.
- You want to host your own models without managing GPUs.
Every Graydient plan has a no-questions-asked 7-day money-back guarantee, so you can test your real workload before committing. You can also cancel, upgrade or downgrade anytime in the billing portal, with prorated charges, so you are never locked into the wrong plan.