Getting started
Novaber workflow: choose, prepare and launch an app
Novaber · Content reviewed
On this page
This guide takes you from a task to an exported result you have inspected. The desktop, Docker, app service and models are separate layers. Opening Novaber proves the workbench can start; it does not establish that every app is ready.
On a new computer, follow installation and setup for drivers, WSL2 and Docker. On a configured computer, check existing components instead of reinstalling them every time.
1. Choose a released app for your task
Select a tool from the catalog on the left and read its capabilities and requirements on the right. Use the app selection guide to compare your input with the output you need.
- For transparent product photos, start with Cutout & Watermark Removal. For scanned text, start with Document Recognition.
- Confirm the media type is supported. Batch images, videos and ordered image sequences can use different workflows.
- Confirm availability. An entry marked in development is a plan; a displayed version does not make it installable.
- Read task-specific restrictions. Voice Studio is currently Beta, and cloning requires authorization for the voice and reference material. Licensing and feature availability are separate questions.
Before continuing: identify the app, supported input and intended export, and confirm the app is released.
2. Check the computer and runtime
For NVIDIA GPU workflows, check driver major >= 575 and Windows driver >= 576.57 for the project’s CUDA 12.9.1 baseline. See the driver and CUDA baseline for version comparison, unknown results and container checks. Check the selected app’s GPU model, VRAM and model requirements separately.
Read system, GPU and storage requirements and the selected app’s requirements. An open workbench or a discrete GPU alone is insufficient.
| Check | Evidence needed | Where to start if it fails |
|---|---|---|
| Windows and WSL | Architecture, system version and virtualization meet current Docker requirements | WSL setup |
| Docker engine | Docker Desktop is running; both Client and Server respond; containers are Linux | Docker verification |
| NVIDIA driver | Full version meets major >= 575 and Windows >= 576.57; unknown versions cannot pass | Driver check |
| Container GPU | Containers can access the required GPU, with enough VRAM for the app | Container GPU check |
| Storage and network | Room for images, extraction, models and exports, with access to required download sources | Storage definitions and network boundaries |
Before continuing: obtain actual check results. A GPU test establishes access for that test; app dependencies, VRAM use and model readiness need their own checks.
3. Install the app and distinguish images from models
- Complete account sign-in and authorization as prompted. These steps and initial downloads need a connection.
- Use the app’s install entry in the workbench and watch download and preparation status. Images prepare the container environment; the desktop installer does not include every app.
- Wait for preparation to finish. If it fails, record the stage and exact message. A failed download does not by itself indicate a GPU fault.
- Check the app’s model instructions. Some weights download on first use. A prepared image does not mean every model has been downloaded.
Preparation depends on image size, network, disk and models; there is no fixed completion time. Preserve errors and existing data before retrying, and read installation troubleshooting.
Before continuing: installation has completed and the interface permits launch or opening the workspace. Resolve downloading, preparing or service-not-ready states first.
4. Open the workspace and check the app service
Use the workbench’s open-workspace entry to reach the local app interface. Each app provides its own task interface; its product description is not a running task.
If it does not open, keeps loading or reports an unavailable service, confirm Docker Desktop is still running and inspect the app status and error. Do not guess a port or change container parameters as a substitute for diagnosis. Record whether the failure occurred after installation or while performing a task, following feedback guidance.
Before continuing: the task interface is reachable without a blocking service or model error. Loading a page alone does not verify inference or export.
5. Verify a small task and its export
Start with a small public or authorized sample before processing a batch. This is a suggested verification method, not a speed or quality guarantee.
- Import supported media and choose a processing mode from the app reference.
- Inspect the preview, text, audio or 3D result instead of relying only on a completed progress bar.
- Export the format you need, locate the saved file and reopen it. Check task completion and export separately.
- Compare with the input. Check transparency for PNG cutouts, text and tables for OCR, playback for speech and the chosen engine’s result for reconstruction.
The batch cutout guide illustrates the interface. Judge model output using your own processed results.
6. Return, update or report a problem
Next time, check Docker and app service status before opening the installed app. Identify desktop and app versions and their official sources separately before updating. Follow download and verification for installer checks, and use the version selector beside an article title to choose its documentation.
After updating a driver, GPU app image or model, recheck the driver baseline, container access and a small task. An old report describes only the previous check. For AI assistance, use the matching versioned setup skill.
Locate the failing stage: environment, image download, model preparation, service startup, processing or export, using the FAQ index if needed. Include versions, the last successful step, exact error and reproduction steps in feedback. If no task was tested, report it as unverified rather than declaring the desktop, container and models all ready.