Latent Realms
Launch app
Blog ·

How Neurons and generation work

Neurons are the compute credits that power every generation on Latent Realms. Here's what a Neuron is, what a generation costs, how the job pipeline runs, and how bringing your own GPU earns free generations.

neuronshow-it-works

Every image you make on Latent Realms is a real compute job running on a real GPU somewhere. Neurons are how we meter that fairly without making you think about instances, VRAM, or hourly rentals.

What a Neuron is

A Neuron is a unit of generation credit. Your account carries a Neuron balance, shown in the top bar. When you run a workflow, the estimated cost is deducted; when it finishes, the actual cost is settled against your balance. You always see an estimate before you commit.

What a generation costs

Cost tracks how much work the GPU actually does — not a flat per-image fee. The main drivers are:

  • Resolution. A 1024×1024 image is cheaper than a 2048×2048 hi-res pass.
  • Steps and sampler. More sampling steps mean more compute.
  • Model size and architecture. A large modern checkpoint costs more per step than a lean SD1.5 model.
  • Extras. Hi-res fix, face detailing, ControlNet passes and upscalers each add work.

Because the estimate is computed from the real graph, turning off a hi-res pass or dropping steps visibly lowers the price before you run.

The job pipeline

When you hit run, this is what happens:

  1. Your graph is validated and normalized into an execution plan.
  2. The plan is dispatched to an available GPU worker.
  3. The worker loads the required model weights (cached when possible), then samples.
  4. Live progress streams back to your browser — steps, previews, and peak VRAM.
  5. The finished image lands in your gallery with its complete recipe attached.

If a model isn't downloaded yet, the worker fetches it from the origin first; that one-time load is why the first generation with a fresh model is slower than the rest.

Bring your own GPU

Latent Realms is also a compute network. Connect your own machine as a worker and your own jobs run free — you're spending your silicon, not our Neurons. Contribute spare capacity to the shared pool and you earn Neurons back when others use it. It's opt-in, and your own work always takes priority on your own hardware.

That's the whole model: predictable credits for cloud generation, free generation on hardware you own, and a way to earn credits by sharing. No subscriptions required to start.


← All posts