Support and troubleshooting
Novaber frequently asked questions
Novaber · Content reviewed
On this page
This page covers general questions about Novaber for Windows. Identify the failing stage, then follow the relevant guide. Use app references for model, format and version details. Desktop, app and setup skill versions are maintained separately.
Locate the failing stage
| What you observe | What to establish first | Continue with |
|---|---|---|
| Unsure whether the computer is suitable | Windows, selected app, GPU, VRAM and actual free storage | System and app requirements |
| Desktop missing or no workspace after launch | Actual installation path, file version and this launch | Setup diagnostics |
| Environment checks fail | Which WSL, local Docker, driver or container GPU check failed | Installation and setup |
| App remains in download or preparation | Images, models or service startup; exact error | Downloads and preparation |
| Workspace opens but processing or export fails | Model, input size, task state and actual files | Tasks and results |
| The cause remains unclear | Last successful step, versions, error and a small reproduction | Feedback guidance |
Novaber and its runtime
What is Novaber?
Novaber is a local AI application workspace for Windows. Browse applications, check your environment, prepare isolated containers, and open local applications from one desktop interface. The catalog always marks which tools are released and which are still in development.
Read the relevant explanation and steps
Who is Novaber for?
Novaber is for creators, engineers, researchers, and small teams that want to run task-focused AI tools on Windows without rebuilding Python, CUDA, and dependency environments for every project.
Read the relevant explanation and steps
Which operating systems are supported?
An x64 installer is currently available for Windows 10 and Windows 11. Check your system before downloading. Support for other operating systems will be announced when available.
Read the relevant explanation and steps
Do I need an NVIDIA GPU?
Requirements depend on the application. Check its preview or detail page for NVIDIA GPU requirements, VRAM, and download size, then run the environment checks. Hardware requirements vary between applications; use the latest catalog entry for the application you plan to run.
Read the relevant explanation and steps
Why is Docker Desktop required?
Novaber prepares isolated container environments for applications. Docker Desktop downloads application images and creates and runs their containers, reducing the need to set up dependencies for each tool manually.
Read the relevant explanation and steps
Do I need an internet connection?
An internet connection is needed to download application images and sign in. Application tasks run on your computer. Fully offline use depends on whether the application needs additional model downloads or external services; local execution does not mean every step is offline.
Read the relevant explanation and steps
How do I install and open an application?
Install Novaber, check Docker, GPU, and VRAM requirements, then choose a tool from the application catalog. Select Install locally, wait for setup to finish, and select Open workspace to enter the application. Initial setup time depends on the image size and your connection.
Read the relevant explanation and steps
Check Windows installer support separately from support in the current Docker Desktop release. A working WSL installation command does not establish that Docker still supports that Windows release. See system requirements and the linked official sources.
GPU, CUDA and Docker
Which NVIDIA driver version is required?
For NVIDIA GPU workflows, check driver major >= 575. The project's CUDA 12.9.1 Windows baseline is 576.57 or newer. A 575.x driver passes the major-version gate but falls below that Windows baseline; 576.02 also fails it.
Read the full version and compare numerically. Unavailable results remain unknown; denied access remains restricted. After the driver passes, verify container access and the app. This is the project baseline, not a universal minimum for CUDA 12.x software. Commands, examples and NVIDIA's corresponding driver versions are in the driver and CUDA baseline.
Does nvidia-smi or its CUDA Version prove the environment is ready?
Check three layers separately:
- Host driver: read its full version and compare with project requirements.
- Container access: follow container GPU validation. A test-image download failure does not establish a GPU failure.
- Actual app: confirm the image, model and VRAM conditions, then process and export a small task.
CUDA Version in nvidia-smi describes driver capability. It does not prove a matching Toolkit is installed or that an app uses CUDA 12.9.1. This Windows + Docker Desktop workflow does not require a complete Windows CUDA Toolkit; do not install a Linux NVIDIA display driver inside WSL.
Why is the environment unready when Docker is installed?
Installed software, a running engine and a ready app service are separate states. Open Docker Desktop, wait for the engine, and confirm the local engine, WSL2 backend and Linux container mode. Interpret checks as follows:
- Only Client output from
docker version: no Server response has been established. linuxfromdocker info --format '{{.OSType}}': the responding engine uses Linux containers; confirm it is the intended local engine.- A responding local Linux engine: continue with GPU, app and model checks before declaring readiness.
Commands and settings are in Docker Desktop preparation. Using Novaber from Windows does not require Docker integration for a particular Ubuntu distribution.
Why do checks fail after enabling WSL?
Enabling Windows features does not establish the WSL version, Linux kernel, hardware virtualization and restart state. Record actual wsl --status and wsl --version output and exact errors, then follow the WSL checks.
A missing version command in older WSL, a timeout or a tool restriction needs further evidence; none automatically means WSL is absent. See restricted and unknown checks.
Installation, downloads and storage
Why are app images and models downloaded after installing Novaber?
The desktop browses, manages and launches apps. Images provide isolated environments; models provide inference capabilities. They are separate resources.
Confirm image download and container preparation, then model readiness. Some weights download on first use. Opening the desktop or finishing app installation does not establish that every model is ready. Follow app and model preparation.
How do I investigate a long download or preparation stage?
Identify image download, extraction, model download or service startup. Record the last changing status and the error that followed, then check:
- Whether Docker Desktop and the local engine are still running.
- Whether the relevant download service is reachable, and whether the error concerns networking, authorization, storage or service startup.
- Free space on the actual Docker, model and temporary-file drives.
There is no universal failure timeout in minutes. Keep the error and follow installation troubleshooting. Clearing Docker data, deleting models or reinstalling everything should not be the first diagnostic step.
Why can preparation fail when the disk appears to have enough space?
Image downloads, unpacked images, model caches, inputs, temporary files and outputs have different sizes. Installing the desktop on one drive does not place all Docker data and models there.
Check the actual failing path and drive, then Docker, model and output locations separately. There is no measured universal total; an installer size or fixed multiplier cannot establish the full requirement. See storage definitions and measurements.
How do I interpret setup checks or locate a custom installation?
Find the failing check, status and reason. Exit code 0 means the report was generated. A pass covers that observation; it does not verify startup or an app task.
Inspect Novaber.exe, its version and required files in the actual directory, then launch that executable. Shortcuts, uninstall records and old logs are supporting evidence; a missing shortcut alone does not call for reinstallation. Follow the file list, status table and launch checks in setup diagnostics.
Tasks, networking and results
Why does a task fail after sign-in?
Check sign-in, app authorization, service startup, models, GPU and input separately. A successful login does not establish all later task requirements, and a task failure does not by itself establish expired membership.
Record whether failure occurs while opening the workspace, downloading a model, starting a task or exporting, with the exact error. Try a small public sample and include versions and stage in feedback.
Does local processing guarantee that data never leaves the computer?
Local execution alone does not provide that guarantee. Sign-in, authorization, updates, images and model downloads may use the network. Check external services and AI clients separately: the inference location and other data flows are different questions.
Before using sensitive material, inspect input/output locations, model downloads and external-interface settings with a public sample. Data and networking describes known boundaries, not a traffic audit or a guarantee of complete offline operation.
Does an open workspace or completed progress mean the output is deliverable?
After establishing page access, confirm the selected model finished computation, the preview belongs to the task, and exported files can be reopened.
Review transparency and edges for cutouts, text and cells for OCR, audio for voice work, and engine-specific reconstruction outputs. Completion, preview, export and quality are separate checks. Follow small-task and export verification.
What should I inspect when processing is slow or results differ from screenshots?
Distinguish model download, waiting for computation, computation and export. Record the model, input size, resolution, GPU and VRAM before comparing times. Tutorial interface examples do not guarantee results for every input.
Keep originals and use representative samples to establish whether the issue repeats. Record input characteristics, method and defect, then consult the app reference. Changing every setting at once makes causes harder to identify.
Can every app shown on the website be installed immediately?
Use the current catalog's release status. Apps in development are not open for use. Published apps still need checks for their current hardware, models and feature scope. See app selection.
Application questions
Where do I investigate cutout, watermark and transparent-export problems?
Distinguish image batches, short MP4 clips and ordered images, then cutout, watermark removal and combined processing. Transparent PNGs, masks and restored images serve different purposes. Use Cutout & Watermark Removal FAQs for diagnosis and the app reference for operations.
Why does a foreground preview not produce a 3D result?
Foreground masks and transparent RGBA are intermediate results. Check camera estimation, sparse structure and the selected MVS, NeRF or 3DGS engine's status and errors. Foreground success does not establish reconstruction success.
Use the 3D reference to check each stage. Report the last successful stage, engine and missing output rather than only saying there is no result.
Why does completed OCR still need review?
Completion does not establish the accuracy of every character, reading order or table. Compare numbers, names, cells and page order with originals, then reopen exports in the intended software.
For errors on particular pages, record input quality, model and mistakes. Follow the OCR reference. Word, Excel or PDF support does not guarantee exact reproduction of every complex layout.
Does SRT import include finished video dubbing?
The current Voice Studio Beta delivers audio, subtitles, role tracks and supporting files. SRT import organizes dialogue and timing; it does not include automatic video dubbing, finished-video export or whole-book assembly.
Review role voices, pronunciation and timing, then finish production in the relevant editor. See the Voice Studio reference for formats and limits.
AI assistance, versions and feedback
Where is the AI setup skill, and how do I choose a version?
The site's current entry follows the selected version. Pin resources with Skill 1.0.3, then use references, diagnostics and the manifest from that package. Do not mix scripts, rules and manifests across versions.
Record desktop, app and skill versions separately. Skill 1.0.3's host driver check does not replace container or task validation; reading the entry does not install software. Follow the AI setup guide.
Can unknown or permission-denied checks be ignored?
unknown means insufficient evidence or a failed query; permission-denied means restricted access. Neither passes or automatically proves a component is missing.
Keep reasons and available evidence. Refer organization policy to the administrator and distinguish AI restrictions from Windows faults. Do not evade explicit restrictions with another script. See diagnostics for permitted observations and skipping restricted WSL checks.
Must I reinstall after a restart or a change of assistant?
Recover the record, then confirm the target computer, pinned release, actual installation and current environment. A historical report describes an earlier observation. Recheck this launch, Docker and unfinished steps; skip components that meet requirements.
Retain the skill version, checksum result, last action, pending restart or permissions and unverified items. Use the handoff template; a full inventory of the computer is unnecessary.
What needs checking after a desktop, driver or app update?
Record what changed and its version. Check startup after desktop updates; driver or GPU app environment changes call for driver, container and small-task checks. Recheck previews and exports after model or output changes.
Match documentation to the app version. An old report does not establish the new state, and task/result migration should not be assumed. Use download verification when checking an installer.
What should I prepare before reporting a problem?
Include desktop and app versions, the skill version for setup issues, Windows and GPU details, the last successful step, exact error and a minimal reproduction. Add full driver and container results for GPU issues; add format, actual files and expected differences for exports.
Remove passwords, tokens, sessions, private paths and unrelated logs. Share only relevant excerpts; private originals are not required. Submission routes and the checklist are in feedback guidance.