LTX Cinescript v1.0.3

Second public build. Still an early beta — please report anything that breaks.

What's new

• Storyboard PDF export works again, and export failures now surface instead of disappearing silently.
• ComfyUI launch respects your own flags, and the app scopes its environment instead of overriding yours.
• Settings and picker fixes carried over from the last build, plus assorted stability work.

Download size — smaller than last time

The installer is ~183 MB. If you already installed v1.0.2, first launch is quick — this build uses the same Python runtime, so there is nothing new to download.

Fresh install? First launch downloads the runtime once: about 3.4 GB over the wire, unpacking to roughly 6.5 GB, with a progress panel. Budget ~15 GB free disk. It resumes if the download drops.

A note on where the runtime comes from

This release deliberately does not carry its own copy of the Python runtime. The app fetches it from the v1.0.2 release assets, because the Python dependencies have not changed — re-uploading 3.4 GB of identical files per release helps nobody.

Practical consequence: the v1.0.2 release must not be deleted or re-tagged while this version is in use. If the dependencies change in a future version, that version will publish its own runtime assets and point at them.

Requirements

• Windows 10 or 11, 64-bit.
• Local generation needs an NVIDIA GPU; VRAM requirements are labeled per model in the app.
• No suitable GPU? The app still works in API mode with your own provider key.

The installer isn't code-signed, so Windows SmartScreen will warn you — "More info" → "Run anyway".

Known limits

• Windows only; on macOS, generation runs through cloud providers.
• Auto-update hasn't been exercised — expect a fresh installer for the next version.

Reporting issues

https://github.com/GenerativeCreations/ltx-cinescript/issues — include what you were doing, local or cloud mode, which model, and logs from %LOCALAPPDATA%\LTXCinescript\logs\.

LTX Cinescript v1.0.2

⚠️ Do not delete or re-tag this release. Later versions (v1.0.3 onward) deliberately fetch their Python runtime from the assets published here, because the dependencies have not changed. Deleting or re-tagging this release breaks first launch for every install that points at it.

---

First public build. This is a tester release — expect rough edges, and please report them.

What it is. A desktop app for turning a script into a shot list and then into video. Write or import a screenplay, it breaks the script into shots, and each shot can be generated as video or images — locally on your own GPU or through a cloud provider.

What's new

• Redesigned model picker — models grouped and filtered by where they run (local or cloud) and what you're making, in plain language instead of raw model IDs.
• One settings surface for API keys, defaults, and local model setup.
• Better ComfyUI support — the app checks that a workflow's models are actually installed, downloads missing ones for you, and the built-in assistant can help set a workflow up.
• More models, including Wan 2.2 image-to-video, a bundled local image model, and OpenRouter support so one key covers many providers.
• A sample script to start from, plus media handling and stability fixes throughout.

Requirements

• Windows 10 or 11, 64-bit.
• Local generation needs an NVIDIA GPU — VRAM requirements are labeled per model in the app.
• No suitable GPU? The app still works in API mode with your own provider key; everything outside generation behaves the same.
• Around 15 GB of free disk for the first launch.

First launch takes a while

The installer is small, but the Python runtime behind it isn't bundled — the app downloads it the first time you run: about 3.4 GB, unpacking to roughly 6.5 GB. Usually 10–30 minutes, with a progress panel. It's a one-time cost per version and it resumes if the download drops.

The installer isn't code-signed, so Windows SmartScreen will warn you — "More info" → "Run anyway". If that's not a trade you want to make, it's a fair reason to skip this build.

Known limits

• Windows only for now; on macOS, generation runs through cloud providers.
• Auto-update hasn't been exercised yet — the next version will likely be a fresh installer.

Reporting issues

https://github.com/GenerativeCreations/ltx-cinescript/issues — helpful to include what you were doing, local or cloud mode, which model, and logs from %LOCALAPPDATA%\LTXCinescript\logs\.