Images, voice and video, on demand.
GPUMax runs four open models on GPUs that wake up when there is work and go back to sleep when there is none. You send a prompt, a picture or a script, and get the file back.
The models
Z-Image Turbo
Fast text-to-image for one-off pictures, thumbnails and batches.
FLUX.2 klein 4B
Keeps the same face across a series from up to four reference pictures, and edits existing images.
Audio8 Air
Text to speech in 23 languages, including Spanish. Clones a voice from a short clip and its transcript.
MiniMax H3
Turns a still picture into a talking or moving clip, lip-synced to your audio track.
How it works
Send a job
Pick a model, write the prompt, and add reference pictures or audio if the model uses them.
The GPU wakes up
The models are already on the card's disk, so a sleeping card is ready in about a minute or two, not fifteen.
It sleeps when idle
After ten minutes with no work the card stops, so idle time costs only its disk.