Skip to content

StoryToolkitAI v0.17.16 - extra alpha
Compare
Choose a tag to compare
@octimot octimot released this 26 Mar 19:31
· 371 commits to main since this release
v0.17.16
73005a6

Download the standalone release here:

StoryToolkitAI 0.17.16.2 for Mac M1, M2 etc.

StoryToolkitAI 0.17.16 for Mac Intel

StoryToolkitAI 0.17.16.2 for Windows (only for CUDA GPUs)

Please report if something doesn't work as expected. So far, only a handful of users have tested this new standalone version!

StoryToolkitAI Patreon

If you find the tool useful, please consider becoming a patron.
This way, we can speed up development to work on cool features that you might need and keep up with rapid advancements in AI.

What's new in this release

Many things have been added and optimized, here are a few highlights:

  • Increased Time Precision improves transcription timings (more updates on this soon)
  • Pre-detect speech to reduce hallucinations on silence and improve speed
  • Support for Whisper large-v2 model
  • Transcription progress now shown in the Transcription Log window
  • Significantly faster Advanced Search when dealing with many transcriptions or files
  • Advanced Search on text files and Resolve markers (via project.json)
  • CMD/CTRL+M in the Transcription Window selects all segments between markers filtered by color or name from Resolve timeline
  • SHIFT+A if text is selected in the Transcription Window will now select all segments under that text
  • Better Find function in Transcription and Advanced Search windows
  • Better handling of Resolve markers
  • many UI and backend optimizations and improvements

For a detailed list of changes since version 0.17.7, see CHANGELOG.

Installation:

macOS

Requirements

macOS 12.6 Monterey or newer
Intel or arm64 (M1, M2 etc.)
8 GB RAM

Installation Steps

Download the zip file that fits with your platform.
Unzip, copy the app file into your Applications folder, then double-click and it should start.

Please also read Info For All Platforms below.

Unsafe warnings

If you get something like "StoryToolkitAI is damaged and can't be opened", try to right click and then click open in the dropdown menu. It will warn you that it's unsafe, but that's just because we haven't notarized the tool with Apple.

If you still have problems opening it, try to take it out of quarantine: open up Terminal and execute:
xattr -d com.apple.quarantine ./StoryToolkitAI.0.17.16.arm64.app (or x86_64 if Intel).
More details on the issue here.

Console Mode

Note: The standalone version does not open up the console. If you're interested to see what is happening behind the hood, open a terminal in the folder where you have the tool (typically in your Downloads or Applications folder) and run this command:

./[STORYTOOLKITAI_APP_NAME].app/Contents/MacOS/[STORYTOOLKIT_APP_NAME]  

# (but replace [STORYTOOLKIT_APP_NAME] with the name of your release)

Windows

Requirements

Windows 10 or newer
CUDA-enabled graphics card

Easy Installation

Just follow the installer to get the tool on your machine.

Notes for Windows

Do not install this version of the tool in Program Files unless you know what you're doing. Some users have experienced permission issues which will make some features unusable.

You don't need to have Python installed on your machine for the standalone version to work.
But, if you have Resolve Studio installed and the tool doesn't start, it's probably because you have multiple Python versions installed on your machine (see #22). This is because Resolve API seems to choose the newest version of Python that is available on the machine, while the tool uses 3.9.13. If you're feeling brave, we recommend installing directly via git and creating an appropriate virtual environment that matches the one that Resolve is using.

Also, if you do need Python for other reasons, make sure you install it from the official Python website and only use other Python installers / versions if you know what you're doing.

If you see an error mentioning python_check, start the tool with the --skip-python-check argument.

Also, see below:

Info for all platforms

Download Permissions

If you get something like PermissionError: [Errno 13] Permission denied: '/Users/[your user]/.cache/torch/hub/trusted_list', it's most likely because the new standalone version uses different ssl certificates (issue #77) , so it's best if you delete the old hub cache, like this:

On macOS

Open terminal, and execute

rm -rf /Users/USERNAME/.cache/hub
rm -rf /Users/USERNAME/.cache/torch
rm -rf /Users/USERNAME/.cache/whisper
  • replace USERNAME with your MacOS user

On Windows

Open CMD, and execute

rmdir /s /q C:\Users\USERNAME\.cache\hub
rmdir /s /q C:\Users\USERNAME\.cache\torch
rmdir /s /q C:\Users\USERNAME\.cache\whisper
  • replace USERNAME with your Windows user

Keep in mind that if you do this, the first time you transcribe or search something it will need to re-download the models, so it will take a bit longer.

Make sure Resolve scripting is enabled

The tool only works with the Studio version of Davinci Resolve.

Davinci Resolve Preferences -> General, "External Scripting using", set to: Local

transcription_WAV

For the optimum transcription quality, try to create the transcription_WAV render preset in Resolve following this info, or simply use WAV audio files or any other format that uses Linear PCM for audio when transcribing - preferably mono.

Downloading models to the cache

The first time you transcribe something with a specific model, it will take a bit longer to start the process because Whisper needs to download the model file (around 1.5GB for the medium model) on your local machine. But, after the model is saved on your machine, transcriptions will take less. The same applies to the advanced search function.

Assets 5