What do you want to see?
Describe a scene, or drop a photo to bring it to life. A realistic clip with sound comes back within 24 hours, for about a cent a second.
Sign in to continue
Your prompt is kept. An account is just an email: it holds your balance and your videos.
At cost, indexed on the GPU market
We add 0.5 % to whatever the cheapest eligible spot GPU costs right now, re-scanned every 5 minutes on two marketplaces. Live: … per second.
Real model, full quality
MiniMax H3, the open-weight model behind the $0.08/s APIs, run with its official 20-step workflow at 1344×768, 24 fps, with native stereo sound.
Nothing to verify
Email or Google to sign in, USDT or USDC to pay, a personal deposit address per account. No card, no KYC, no subscription.
What a second of video costs
Full-quality MiniMax H3 at 768p, price per second of generated video. Ours is live, the others are their published pay-as-you-go rates.
Price per second of video, USD
Same model everywhere it is offered. Hover a bar for the price of a 10-second clip.
By delivery window
The longer you can wait, the deeper we batch, the less you pay. Locked when you submit, refunded if the job fails.
| Model | Within 24 h | Within 4 h | Within 1 h | Typical API |
|---|
Made for the everyday shot
Product b-roll, reels, ad variants, dataset generation. Anything where waiting a few hours is worth paying ten times less.
How it works
1 · You submit
Text or a photo, a length, a deadline. The price is computed from the live GPU market and locked for your job.
2 · We wait for the dip
A scheduler watches interruptible GPU offers. When enough work is queued and a GPU is cheap, it rents it, loads the model once and renders the batch back to back.
3 · Your clip lands here
It appears in your account, and hits your webhook if you use the API. Missed deadlines are rerun on on-demand GPUs at our cost.
Questions
Why is it this cheap?
Real-time APIs keep GPUs warm around the clock and bill you for the idle time. We rent interruptible GPUs only while a batch renders, pick the cheapest one that can run the model at that moment, and add half a percent. Your patience is the discount.
Does it need a starting image?
No. Describe the scene and the model creates everything. Drop a photo when you want to keep a specific face, product or place: it becomes the first frame.
What is draft mode?
The same model with the official turbo LoRA in 8 steps instead of 20. Much cheaper, slightly softer motion. Good for iterating on a prompt before the final render.
How do I pay?
USDT or USDC on BNB Smart Chain (BEP-20) to your personal deposit address, by QR code or copy-paste. Balance credited in about a minute. No card, no KYC. Card payments are coming later at +10 %.
What is not allowed?
Two things, and only two: sexual content involving minors, and non-consensual sexual content. Those prompts are refused before anything is charged. Everything else is passed to the model as you wrote it.
Can I automate it?
Yes. One endpoint to submit, one to poll, a webhook on completion. See the API.
Is this a beta?
Yes. Capacity is booked on demand, so early jobs may take the full 24 h. Anything that fails is refunded.