GPU cost calculator
Estimate the GPU rental cost of one SadTalker render or a batch. Enter your hourly rate, choose a recorded render time or enter your own, and set the video count.
Runs in your browser. No uploads, GPU connection or live rental-price lookup.
How the estimate works
Cost per video = hourly price × render seconds ÷ 3,600. Batch cost = cost per video × number of videos. Batch render time = render seconds × number of videos.
For example, at an illustrative $0.50/hour, a 60-second render costs $0.0083. A batch of 100 costs $0.8333 and takes 1 hour 40 minutes of rendering. This example is not a current rental quote.
What your rental bill can include
This estimate covers the entered render time only. A rented instance may keep charging while you install dependencies, download models, wait between runs or retry failures. Storage, bandwidth, taxes and minimum billing increments are excluded. Check the terms of your rental and stop the instance when finished.
The October ComfyUI presets include input loading, worker/model startup, inference and export during each measured ComfyUI execution. They exclude queue wait and initial setup. Do not add those included stages twice. Repeat your own timing before estimating a large batch; different audio lengths, settings and hardware can change the result.
Recorded times and setup guides
- 256 vs 512: measured render times
- GFPGAN ON vs OFF: added processing time
- GPU benchmark comparison
- Installation guide
- Cloud and Vast setup guide
The October ComfyUI measurements use a different setup from the earlier GPU comparison. The RTX 3090 preset comes from the earlier GPU comparison; the four October RTX 4090 presets come from the ComfyUI test data. Each preset identifies its source setup.