Voice Designer / Preview user guide

Requirements & troubleshooting.For the unsigned preview.

Hardware guidance and symptom-first troubleshooting for the free unsigned Voice Designer 0.1.6 preview. These numbers describe a current experimental voice-generator path and provisional planning targets—not a certified minimum for the existing one-off unsigned preview.

Final and only planned release: 0.1.6 · Free unsigned preview · No later final build
Cinematic future Voice Designer interface with microphone and holographic waveforms, not an operational screenshot
Proposed interface illustration · preview download exists, but the native GUI is not fully tested

Oct 25 update · actual tests and limits

0.1.6: backend worked, graphical controls still untested.

Version policy: 0.1.6 is the final and only planned standalone release of the older Voice Designer generator. Its already downloadable unsigned preview is that release, not a stage before a separate final build. Kira World instead plans a virtual Voice Box, still under development. The present preview limitations remain important. On the existing Windows 11 25H2 computer, the unsigned preview successfully installed an isolated Python runtime, dependencies and a local model. Eleven of eleven model files were checked, and two different original synthetic voices were generated by the installed command-line backend: 2.72 and 3.92 seconds. Uninstall and reinstall passed, with saved WAV and cached model data retained.

Limitations: the new native interface has not been opened or button-tested. A clean Windows computer, missing Microsoft C++ runtime, lower RAM/VRAM, and non-NVIDIA GPU routes are unqualified. The successful test computer had 32 GB RAM and an RTX 5060 Ti with 16 GB dedicated VRAM, not certified minimum specifications.

Download the unsigned preview and see warnings → · Read the test results →

Computer guidance · draft

Experimental checks versus planning recommendations

The middle column describes current engineering guards, not universal minimum requirements. The right column contains suggested planning targets that must be tested before publishing a supported-computer list.

ComponentCurrent experimental statusProvisional planning target
WindowsCurrent wrapper uses 64-bit Windows. Supported editions are not yet certified.64-bit Windows 11, subject to final installer and driver qualification.
CPUThe experiment has used two logical processors; no measured minimum CPU model is established.Modern x64 CPU with four or more cores.
System RAMThe experimental preflight asks for roughly 10 GiB available before startup, including a 6 GiB voice-family limit and 3 GiB free-memory reserve. Not an installed-RAM minimum.32 GB installed, with roughly 16 GiB available before running the app.
GPU & VRAMCurrent configured CUDA path checks around 9 GiB total and free VRAM before loading, and reserves an additional 1 GiB. GPU fit has not yet been proven.Compatible NVIDIA GPU with 16 GB dedicated VRAM; 12 GiB free is a planning target, not a certified requirement.
DiskThe selected cached Qwen payload alone is approximately 4.52 GB (4.21 GiB). It excludes dependencies, additional models, generated audio and temporary files.SSD with about 30 GB free as a planning allowance; final requirements depend on the actual release package.
DriversThe NVIDIA driver must support the CUDA/PyTorch runtime selected in the release.Use officially distributed drivers tested with the final package; no single version floor is certified.
InternetInitial model and runtime downloads require a connection. Offline operation requires a complete, compatible local cache.Stable connection for first setup, followed by a successful disconnected-computer test before any unconditional offline promise.
AudioA working Windows playback device is needed to hear generated previews; optional authorized reference capture requires a microphone.Headphones for careful review and a quiet recording setup when supplying your own authorized material.

Units: GB is decimal; GiB is binary. Available RAM differs from installed RAM, and shared GPU memory is not equivalent to dedicated VRAM. Smaller engines and CPU fallbacks each require independent validation. The table is not a purchase recommendation.

Troubleshooting · keep the useful error

Start with what went wrong.

01

Installer will not start

Check whether the existing unsigned 0.1.6 preview supports your Windows edition and CPU architecture. Record the full error; do not disable system protection to force an unverified installer.

02

No GPU or CUDA available

Check the selected engine's supported GPU, driver and CUDA/PyTorch combination. Integrated or non-NVIDIA graphics do not satisfy the current CUDA route. No untested CPU fallback is promised.

03

Insufficient RAM or VRAM

Save other work, close unnecessary applications and re-check *available* memory. Do not disable safety guards or assume that installed capacity equals free capacity.

04

Driver or DLL error

Use the release's reviewed runtime and official vendor/Windows driver updates. Avoid downloading individual DLL files from unofficial websites.

05

Interrupted download or missing model

Check free disk space, internet connection and the future documented repair flow. Incomplete files must fail integrity verification.

06

App opens but generation fails

Preserve a sanitized full error and the selected environment. The present internal Windows path-check failure occurred before audio generation; it is not an audio-quality finding.

07

No sound

Verify that a nonempty WAV was produced and can be played outside the app. Then review Windows output device, volume and mute settings. A silent mock is not proof of synthesis.

08

Audio distorted or words missing

Keep the failed WAV and settings. Repeat with shorter, clearly punctuated text only using supported controls; never overwrite a previously approved voice reference.

09

Chatterbox output sounds different

Confirm which authorized reference WAV was used. Compare the same spoken passages and retain both outputs. Similarity across engines requires separate tests.

10

Offline setup fails or disk is full

Complete and verify the entire model cache while connected. Leave room for dependencies, temporary audio and backups before testing offline operation.

Support request checklist

What to keep when asking for help.

Record the app and model versions, Windows edition, CPU, installed and available RAM, GPU and dedicated VRAM, driver version, chosen engine, the precise failed step and a sanitized error log. State whether a WAV was produced and whether you could hear it. Remove passwords, tokens, private recordings and personal file paths before sharing.

October 25 source evidence establishes backend generation and install/reinstall on one Windows computer, but not tested 0.1.6 native GUI controls or a qualified general-public installer. The requirements remain provisional.