Stability issues with Local AI, Gemini API keys, LM Studio and UI + workflow suggestions

[ericwot3-sys]

Hi,

First of all, thank you for WorldScript Studio. I really like the concept and the amount of features already available. I have been testing the desktop application and noticed several bugs and possible improvements.

1. Local AI model download freezes

In the Local AI download section, when I start downloading a model, the interface often freezes.

The progress bar does not visibly move and sometimes nothing changes for a very long time. I waited approximately 20 minutes once without any visible progress.

It would be useful to have:

• a reliable progress percentage
• downloaded MB / total MB
• current download speed
• cancel/retry buttons
• a clear error message if the download fails

2. General UI freezes / performance

The application occasionally freezes or becomes noticeably slow.

I am running it on a gaming PC with 32 GB of RAM and a dedicated GPU, so the interface should normally remain responsive.

Scrolling through some sections can also cause visual elements or text to overlap other panels/tabs.

3. Gemini API key validation appears outdated

The application seems to expect Gemini API keys beginning with the older AIza... format.

Google AI Studio is now issuing newer authentication keys beginning with AQ....

Please consider updating the Gemini key validation so it does not reject legitimate newer keys based only on their prefix.

4. LM Studio integration

I connected LM Studio, but WorldScript Studio sometimes returns connection/error codes even though the LM Studio server is configured and running.

Better diagnostics would help here, for example:

• server reachable / unreachable
• detected models
• endpoint being used
• connection test
• exact error details

5. Python detection

Python is installed correctly on my computer, but WorldScript Studio sometimes reports that Python is not installed.

It would be helpful if the application displayed:

• detected Python executable path
• detected Python version
• expected/minimum version
• a button to manually select the Python executable

6. Dedicated “Scenario / Screenplay” workspace

I would really like a dedicated Scenario / Screenplay section in the sidebar.

The idea would be to bring together the information necessary for screenwriting in one workspace:

• Characters
• Universe / world-building
• Pitch / logline
• Synopsis
• Story structure
• Scenes
• Notes
• Storyboard
• Manuscript / screenplay

A workflow closer to a structured Google Docs writing workspace could make larger audiovisual projects much easier to understand and organise.

7. Writing interface / background

The writing interface currently has background patterns / small orange marks that can become visually tiring.

It would be nice to have a cleaner writing mode with:

• plain background
• additional neutral themes
• more appearance/customisation options
• possibility to completely disable background textures and decorative patterns

8. Plugins / extensions

A plugin or extension system could be very interesting in the future, allowing the community to add tools without making the core application too complicated.

9. Real-time collaboration

A collaborative document mode would also be extremely useful.

For example, from a project/document, a user could generate an invitation URL with permissions and an expiration time such as:

• 10 days
• 15 days
• 30 days
• no expiration

Another person could open the link from their computer and work on the same project in real time, similar to Google Docs.

Overall, I think WorldScript Studio already has a lot of potential and many useful features. My main request would be to prioritise stability and interface optimisation before adding too many additional features, because occasional freezes and visual glitches currently make longer writing sessions difficult.

Thanks for your work and I hope this feedback can help improve the application.

[qnbs]

Thank you for the detailed report — going through each item individually against current `main`, not just the prior PR description, per your ask to separate what's actually fixed from what's still open or a feature request.

Already fixed (merged in #336, commit 78c7bb7)

• Item 3 (Gemini key validation) — the brittle `AIza`-prefix rejection is gone; validation now does minimal syntax checks plus real provider validation, so newer `AQ...`-prefixed keys are accepted.
• Item 4 (LM Studio diagnostics) — LM Studio discovery/testing now uses the correct OpenAI-compatible `/v1/models` endpoint, with sanitized diagnostics (reachability, detected models, endpoint, connection test, error details) surfaced in the UI.
• Item 5 (Python detection) — desktop Python discovery, executable selection, and version/error reporting were centralized; detected path/version and manual-selection UX are in place.
• Item 7 (writing interface background) — a plain writing-surface style already exists (`writingSurfaceStyle: 'textured' | 'plain'`) alongside the textured default, letting you fully disable the decorative background pattern.

Fixed in this round

• Item 1 (download progress) — fix(ai): download progress bytes/speed + voice progress-scale bug (#333 item 1) #346 fixes a real bug where the voice-model download progress bar looked frozen for most of the download (a 0-100-vs-0-1 scale mismatch that clamped it at ~95% almost immediately), and adds downloaded/total MB + live speed to both the voice-model and local-AI-text-model download UIs. The WebLLM text-model byte counts are an approximation (the underlying library doesn't expose real byte-level progress) — clearly labeled as such in the UI, not presented as exact.
• Item 2 (general UI freezes) — shares its root cause with [Bug]: Sluggishness? (.deb) #332: fix(desktop): Tauri boot hydration, flush-on-quit, and Settings re-render perf (#332) #345 fixes a desktop boot-hydration bug (state was never read back on launch), a Settings-panel unnecessary-re-render bug, and extends backdrop-blur GPU-cost mitigation. See the [Bug]: Sluggishness? (.deb) #332 comment for the full writeup. As with [Bug]: Sluggishness? (.deb) #332, I can't run a packaged `.deb` build here to confirm this fully resolves the freezes you saw — re-testing on your hardware once fix(desktop): Tauri boot hydration, flush-on-quit, and Settings re-render perf (#332) #345 merges would help close this out.

Split into standalone feature requests

These are genuine new-feature asks, not bugs — tracked separately so they don't keep this issue open indefinitely as a catch-all:

• Item 6 (dedicated Scenario/Screenplay workspace) → Dedicated Scenario / Screenplay workspace #347
• Item 8 (plugin install/marketplace UI — the sandboxed execution engine already exists, only end-user installability is missing) → Plugin install / marketplace UI #348
• Item 9 (collaboration invite links with expiry/permissions — real-time collaboration already exists, only the invite-link UX is missing) → Collaboration invite links with expiry/permissions #349

Leaving this issue open for item 2's packaged-build confirmation; everything else above is either already resolved or tracked in its own issue now.