What we build
We build the part that remembers.
Vayra is a text-to-image engine built around one rule: nothing leaves the queue without the settings that made it. The picture is the easy half. Getting that same picture again in March is the half that decides whether a team can put this in a workflow.
So the console shows the settings before it shows the render. A frame from one project forks into another without retyping anything, because the file already carries how it was made. The result is a record your team can reopen, audit and reuse.
- Seed
- 4471820
- Sampler
- dpm++ 2m karras
- Steps
- 34
- Model
- v3.1
- Made
- 2026-03-04
How it is built
Reproducible by default
The seed lands in the file the moment the render does, next to the model version and the date. No panel to open. Nothing to remember to save.
Batch first, then pick
Single renders are a habit from slower tools. The queue runs one prompt across a spread of variations, retries the ones that fail, and bills only the frames that finish.
Grayscale on purpose
The interface has no color of its own. Every hue on screen comes out of the model, so a render that looks warm is warm, and not the panel sitting behind it.
Readable output
PNG with a sidecar JSON, plus a CSV for the batch. The metadata opens in a text editor, so the record outlives whatever you view it in.
Limits
Things it will not do.
There is no video here, and no in-browser painting. Both would have slowed the queue for people who only need frames, and the queue is the thing we tune first.
It also assumes you write your own prompts. The console suggests nothing and rewrites nothing. That is a real limit, and it is also why nothing here nudges every user toward the same look.