Cost and speed
Cobalt Pear is free. You pay RunPod for the GPU, by the second, only while a worker is running. When no picture is being made, the worker turns off and the GPU costs nothing.
What a picture costs
Section titled “What a picture costs”The seconds the GPU works × RunPod’s price for that GPU — see runpod.io/pricing (Serverless). Your RunPod Billing page shows what you really paid.
Waiting for the first picture
Section titled “Waiting for the first picture”- After a build, the very first picture waits while the worker fetches its image and starts. This is the longest wait.
- After the worker has been off, the next picture waits for it to start again — usually much shorter. FlashBoot on makes it shorter still.
- Pictures right after each other skip that and come back fastest.
What makes it slower or dearer
Section titled “What makes it slower or dearer”- Bigger GPU. A model that needs 48 GB costs more per second than one that fits in 24 GB.
- Bigger pictures and more steps take more seconds each.
- LoRAs by link are downloaded again each time the worker starts (LoRAs).
Paying nothing while idle
Section titled “Paying nothing while idle”A worker turns off a few seconds after its last picture, so an endpoint you aren’t using costs nothing. Don’t set Active workers above 0 on RunPod — an active worker stays on, and you pay for it all the time.