flet build macos now supports NumPy (official bug-fix)

There was a bug in Flet (version 0.21.0 and 0.21.1) that caused built app with NumPy for macOS to crash immediately after launch. I reported this issue on GitHub some time ago, and finally Feodor Fitsner, the author, updated me with a solution. To fix the issue, you simply need to build your Flet app again.


Template for Flet version 0.21.2 and the latest version 0.22.0 (as of April 19, 2024) has been modified, according to the author. Flet itself has not been changed. Therefore, if you are using NumPy with a Flet app on macOS, you should be good to go with these versions. Just rebuild now (flet build macos) and it should work.

If you’re running older version of Flet, upgrade to the latest with pip install --upgrade flet. You can specify a version as well if needed:

pip install flet==0.21.2

I upgraded to 0.22.0 and confirmed my SPS Player app built for macOS with NumPy code didn’t crash by building it again. If not sure your Flet version, try the below command:

pip list|grep flet

# result will be like the below:
flet                  0.21.2
flet-core             0.21.2
flet-runtime          0.21.2

Another way

Adding the below code to your Flet will also resolves the issue, according to the comment in the issue. I won’t bother as long as my codes work with the recent Flet versions though.

import os

os.environ["OPENBLAS_NUM_THREADS"] = "1"

In the related issue you can find more, detailed info.

Thanks to the solution

Flet is open-source software that is available for free, so I didn’t request for update or pressure the author to fix anything. However, I am grateful that they took the time to address this issue and make Flet even better. To contribute to Flet, I continue developing my app with it.

Image by Stable Diffusion

You can see what I wanted from the prompt. The quality of each blog post’s image is vastly different and it’s getting out of hand. Since I got the M2 Max, I’ve been asking for 6 images with 30 steps and then adjusting the step and prompt or model based on the outcome.

2024-4-20 0:14:30


512 x 512

Include in Image:
fantazy, realistic painting, a wizard with a magic wand killed misterious creature by fireball

Exclude from Image:



Guidance Scale:


ML Compute Unit:

Successful build of Flet app with Audio control

I have successfully built my own app, Speech + Subtitles Player made with Flet for Python by eliminating a line of code that required NumPy. However, adding copyright by template caused building process to fail probably because of the audio control which seems to conflict with template file. Unfortunately, the official document does not mention challenges about NumPy or Audio + template on macOS. If you’re looking for the solution, you found the right place.

(My) Environment

The issues and the solutions were found with the followings. Different versions of Flet and/or Flutter might result something different.

  • Hardware: Mac Studio M2 Max 12-core CPU / 30-core GPU / 32GB RAM
  • OS: macOS Sonoma 14.3.1
  • Xcode: 15.3
  • Python 3.11.7
  • Flet: 0.21.2
  • Flutter 3.19.3
  • Dart: 3.3.1
  • CocoaPods: 1.15.2

Rewrite code to eliminate NumPy

My previous code used NumPy only to find the closed value of a subtitle’s end time to scroll based on the current position of audio. It’s line #520. By changing it to the line #522, I was able to remove NumPy from my code (GitHub code is updated). Commenting # import numpy as np successfuly built the app by executing flet build macos --include-packages flet_audio now builds app with no issue. I do not notice any delay by making this change.

    # Called when slider position is changed and scroll to subtitle with the nearest end_time.
    async def scroll_to(self, e):
        end_time = [item[2] for item in self.subtitles]
        # Numpy is only used below:
        #index = np.argmin(np.abs(np.array(end_time) - e))
        # Below works without using Numpy:
        index = min(range(len(end_time)), key=lambda i: abs(end_time[i]-e))
        self.subs_view.scroll_to(key=key, duration =1000)

Should you use NumPy a lot for a time sensitive app, you need to wait for Flet (flet build) to work with NumPy on Mac (I submitted an issue ). Alternatively, you can use flet pack main.py which is not suggested officially but works (see the official document).

Simple build works but it fails by specifying template

As a next step, I tried to add copyright by specifying the template directory and faced another issue below. It shows lots of complicated version dependencies.

% flet build macos --build-version "1.0.1" --template flet-build-template --include-packages flet_audio
Creating Flutter bootstrap project…OK
Customizing app icons and splash images…OK
Generating app icons…Because flet_audio <0.20.1 depends on flet ^0.20.0 and flet_audio >=0.20.1 <0.20.2 depends on flet ^0.20.1, flet_audio <0.20.2 requires flet ^0.20.0. 
And because flet_audio ^0.20.2 depends on flet ^0.20.2 and flet_audio >=0.21.0 <0.21.1 depends on flet ^0.21.0, flet_audio <0.21.1 requires flet ^0.20.0 or ^0.21.0. 
And because flet_audio >=0.21.1 <0.21.2 depends on flet ^0.21.1 and flet_audio >=0.21.2 depends on flet ^0.21.2, every version of flet_audio requires flet ^0.20.0 or >=0.21.0 <0.22.0.
So, because fletaudioplayback depends on both flet ^0.19.0 and flet_audio any, version solving failed.

You can try the following suggestion to make the pubspec resolve:
* Try upgrading your constraint on flet: dart pub add flet:^0.21.2

Error building Flet app - see the log of failed command above.

To avoid this error, you must forget about the template.

SOLUTION: Use command options instead of template

I know it’s not smart but that’s the only way I found. Trash the template folder and use a couple of options as blow. Hope audio + template will work in the near future as well.

flet build macos --build-version "1.0.1" \
--copyright "Copyright (c) 2024 Peddals.com" \
--product "Speech+SubtitlesPlayer" --include-packages flet_audio

Quick hints: --product can name your app. If you don’t use this option, parent folder name will be your app name. --include-packages flet_audio is a requirement for an app that uses the Audio control. See this page for the full list of options.

Additional tip regarding your folder structure — when you build an app, remove all unnecessary files and folders. When I first built this app with few large audio files which were in another folder than the assets folder, building process took a long time and built app was 1.7GB in size. After removing the audio files, the app size became 260MB.

Some notes about the built app

When I launched the app in build/macos first time, I saw a white blank window and it took several seconds to draw contents such as buttons, texts, and a logo image. The reason was not the app itself, but an anti-virus software on my Mac that was running so hard consuming 300% of CPU to scan the app. Once it’s done, the app started running quickly. While playing audio, CPU usage is around 30-40%. I’m not sure why but the splash logo does not appear when the app is launched. Since it does not harm functionalities, I decided not to bother. Everything else works as expected. Nice.

Bonus: Get coding assistance from Copilot (Bing Chat)

To rewrite the function with NumPy to native Python code, I got an assistance from Copilot, formerly Bing Chat free version. You better use paid version of generative AI services for serious production for sure, but the fee version works fine just as long as you understand that a few chat back and forth is needed to get an expected answer. Try giving enough but short requirements and complete in 5 round-trips. Below is an example, actual chat I had with Copilot.


can you help me rewrite an existing python code with numpy to python native without numpy? I’ll share a code with you


Certainly! Please go ahead and share the existing Python code that uses NumPy, and I’ll help you rewrite it using native Python constructs. 😊


end_time is a list containing integers from the smallest to the largest. e is also an integer. import numpy as np. This is the actual code to be rewritten: index = np.argmin(np.abs(np.array(end_time) – e))


Certainly! Let’s rewrite the given code snippet that uses NumPy to achieve the same functionality using native Python without relying on NumPy.

The original code snippet calculates the index of the element in the end_time list that is closest to the value of e. We’ll achieve the same result using native Python constructs.

Here’s the rewritten code:

def find_closest_index(end_time, e):
    # Initialize variables to keep track of the minimum difference and index
    min_difference = float('inf')  # Initialize with positive infinity
    closest_index = None

    # Iterate through the end_time list
    for i, time_value in enumerate(end_time):
        # Calculate the absolute difference between the current value and e
        difference = abs(time_value - e)

        # Update the closest index if the current difference is smaller
        if difference < min_difference:
            min_difference = difference
            closest_index = i

    return closest_index

# Example usage
end_time = [1, 3, 6, 9, 12]  # Replace with your actual end_time list
e = 7  # Replace with your actual value of e
result_index = find_closest_index(end_time, e)
print("Closest index:", result_index)

In this rewritten code:

  • We manually iterate through the end_time list.
  • We calculate the absolute difference between each value in the list and e.
  • We keep track of the index corresponding to the smallest difference encountered.

Feel free to adapt this approach to your specific use case, and let me know if you need further assistance! 😊


can you rewrite it to a one-liner with lambda?


Certainly! Here’s the equivalent one-liner using a lambda function:

find_closest_index = lambda end_time, e: min(range(len(end_time)), key=lambda i: abs(end_time[i] - e))

This lambda function calculates the index of the element in the end_time list that is closest to the value of e. It uses the min() function with a custom key function to find the index with the smallest absolute difference.

Feel free to use this concise one-liner in your code! 😊

Image by Stable Diffusion

This time I needed to exclude few words as initially generated images were all fat old man with big stomach. Face in close-up is detailed enough, but when it’s far Stable Diffusion tends to lose focus. Thumbnail looks fine though.

2024-3-31 1:55:34


512 x 512

Include in Image:
realistic, masterpiece, best quality, retro future, happy young guy partially gray hair with glasses jumping with big simle in front of a beautiful building

Exclude from Image:
frame, old, fat, suit



Guidance Scale:


ML Compute Unit:
CPU & Neural Engine

Speech + Subtitles (SRT) Player app made with Flet

If you have not, try OpenAI’s Whisper, speech-to-text AI. It’s free and works on a personal computer without sending data to the cloud. With my M1 mac mini 16GB RAM, it takes about 90% of the actual length of audio file to complete the transcription, and the accuracy/quality is also about 90% in Japanese or even better in English. After transcribing multiple types of speeches, I researched and tried various parameter changes to increase the accuracy, but it turned out to be nearly impossible, at least for now. So, I changed my mind and started building an app that I can edit subtitles easily while playing speech audio file simultaneously. In this post, I’m going to introduce the standalone desktop app that I built with Flet, GUI application development framework for Python. l’m going to write another post to share more about the code. For the time being, you can read my code that I added comments as many as I could (GitHub).


I call the app Speech plus Subtitles Player (SPSP or SPS Player) in English. I named it Jimaku Gokuraku Maru in Japanese (“jimaku” means subtitles), named after an old console game for Famicom, Jigoku Gokuraku Maru (“jigoku” means hell. haha). As my app is an open source software it really does not matter, but I also created a cool logo using a free Japanese font Gen Kai Min Cho from Flop Design. Anyways, with SPSP, you can play audio file and edit subtitles (SRT file) generated by Whisper or any other speech-to-text tool quite easily. Subtitles scroll in sync with the audio, and the UI is intuitive.

Needed features are all included. Subtitles above are actual Whisper output of a famous speech by Steve Jobs.

How to use SPSP

(See “How to run SPSP” below to run the app.) When you load an audio file using the Open Speech File button, if a subtitle file (extension .srt) or text (extension .txt) with the same name exists in the same folder, it will be automatically opened. The Play button plays/pauses the audio and the subtitles (SRT file) scroll along with the audio. Click on the timestamp to cue there. Click on the subtitle area to edit the text. The edited texts will be overwritten in the same file by clicking on the Save button. The SRT and TXT buttons will exported as separate files.  The 1.5x and Auto scroll switches allow you to turn on/off 1.5x faster playback speed and automatic scrolling of subtitles (auto scrolling is only available for SRT format). There is currently a known issue where clicking the Open/Export as button does not open the dialog and the entire app stops working sometimes for some reason. Please be sure to click Save frequently when you’re editing.

When exporting as TXT, it does not include timestamps, so auto-scrolling within the app is not possible, but it is useful for various purposes such as meeting minutes and reports. SRT is a popular subtitles format (Wikipedia), and if the original audio data is a video, it can be imported as subtitles data using video editor software such as DaVinci Resolve.

Target users and use cases

The main target users are those who want to embed subtitles in own videos, typically YouTubers. Also, SPSP is useful for engineers who verify the accuracy of transcription AI, including Whisper, and operators who write conversation reports at call centers. In addition, it can be useful for meeting minutes, or for learning foreign languages (despite the accuracy varies, Whisper supports quite a few languages: Supported languages ​​).

How to run SPSP

Although I say it’s an “app,” it is currently not possible to build as executable that opens by double-click. You need to set up an environment and run a command to launch SPSP. I figured out how to avoid crashing and errors to build this app. Please see my another post if you wan to build. If you are using Windows or Linux, you may need to take additional steps, so please refer to the Flet official documentation. The code of SPSP is on GitHub:


Clone the code, create a Python virtual environment, and install Flet and Numpy

Python version has to be 3.8 or newer (mine is 3.11.7) The example below specifies python 3.11 and uses pipenv, but any virtual environment is fine.

git clone https://github.com/tokyohandsome/Speech-plus-Subtitles-Player.git
cd Speech-plus-Subtitles-Player
pipenv --python 3.11
pipenv shell
pip install flet
pip install numpy

Run the app

Once the environment is created, you can run the app SPSP with the command below.

python main.py

Select audio file

After launching, click the Open Speech File button and select an audio file such as MP3 or WAV. The first time on macOS, you will be asked if you want to give access to your Documents folder, so please approve. If there is a file with the same file name and .srt (or .txt) extension in the same folder, it will be loaded automatically. You can also manually load subtitles file after loading an audio file.

Known issues and limitations

I don’t think there are any critical issues, but just in case, you might want to keep copy of subtitles file in a different location before opening in SPSP. Here are some known issues and limitations:

  • If there are many subtitle buttons, it will stutter when moving or resizing the window.
  • When you build app by flet build macos --include-packages flet_audio, built app crashes (Flet version == 0.21.2). If you want an executable and don’t need auto-scroll, comment out the line import numpy as np. –> This is now resolved by eliminating NumPy.
  • Sometimes when you click the Open or Export button, the dialog will not open and you are unable to do anything other than closing the app. We are currently investigating the cause. Please save frequently.
  • It seems that the sample rate of MP3 that can be played on macOS is up to 44.1KHz. If it is higher than that, please convert by using Audacity etc.
  • Add audio file extension to the pick_speech_file method if it’s grayed out.
  • SRT format originally seems to allow multiple-subtitle lines per block, but this app only expects 1 line. There should be no problem with SRT files exported with Whisper.


I won’t go into details, but here are how to download online videos such as YouTube as audio files, and how to transcribe to SRT using mlx version of Whisper, which Apple has optimized for macOS.

Download online video as m4a audio file

ffmpeg will be installed system-wide. You better create a dedicated virtual environment.

brew install ffmpeg
pip install yt_dlp
python -m yt_dlp -f 140 "url_of_online_video"

Python script to transcribe audio file to SRT with Whisper

For macOS, create an environment where MLX version of Whisper can run, download whisper-large-v3-mlx from Hugging Face, and place json and npz files in mlx_models/whisper-large-v3-mlx folder. Then create the speech2srt.py file (below). Edit path_to_the_folder, audio_file_name and language='en' to meet your file/language. If you set a different language like ja, Whisper tries to transcribe and translate into Japanese, but quality is not good.

import whisper
import time
import os

base_dir = "path_to_the_folder"
speech_file_name = "audio_file_name"

start_time = time.time()
speech_file = base_dir + speech_file_name
model = "mlx_models/whisper-large-v3-mlx" 

result = whisper.transcribe(
                            temperature=(0.0, 0.2, 0.4, 0.6, 0.8, 1.0),

end_time = time.time()
elapsed_time = round(end_time - start_time, 1)

print(f"Time elapsed: {elapsed_time} seconds")

def ms_to_srt_time(milliseconds):
    seconds = int(milliseconds / 1000)
    h = seconds // 3600
    m = (seconds - h * 3600) // 60
    s = seconds - h * 3600 - m * 60
    n = round(milliseconds % 1000)
    return f"{h:02}:{m:02}:{s:02},{n:03}"

subs = []
sub = []
for i in range(len(result["segments"])):
    start_time = ms_to_srt_time(result["segments"][i]["start"]*1000)
    end_time = ms_to_srt_time(result["segments"][i]["end"]*1000)
    text = result["segments"][i]["text"]

    sub = [str(i+1), start_time+' --> '+end_time, text+'n']

text_file = base_dir + os.path.splitext(os.path.basename(speech_file_name))[0] + ".srt"

# Overwrites file if exists.
with open(text_file, 'w') as txt:
    for i in subs:
        for j in range(len(i)):
            txt.write('%sn' % i[j])

Now run as below and an SRT file will be created in the same folder as the audio file. Please note that existing SRT file will be overwritten.

python speech2srt.py

MLX Whisper uses GPU. My M2 Max Mac Studio (30 core GPU) completes transcription about 1/6 of the audio length.

Image by Stable Diffusion

Increasing steps didn’t help fix scary fingers, also the good taste of both lady’s good faces focusing on speeches to transcribe couldn’t be kept. Larger steps does not always mean better quality or taste.

March 24, 2024 23:45:42


512 x 512

Include in Image:
realistic, masterpiece, best quality, retro future, office ladies transcribing audio from record player

Exclude from Image:



Guidance Scale:


ML Compute Unit:
CPU & Neural Engine

Deploy client-side (static) web app built by Flet

I edited width, alignment, etc. of my Flet code originally written as a desktop app to make it work as a web app. In this post I’m going to introduce how to build and deploy your Flet app as a static website (client-side, HTML + JavaScript). This deployment should work on an ordinary web hosting server, and you don’t need to know about web server side technology. You can learn how to add Google AdSense advertisement to your Flet web app as well.


Please refer to my previous post and build your environment. It’s mainly targeting macOS. The sample Flet app code is used in this post.

How to build Python GUI app on macOS with Flet

In case you’re looking for server side deployment

My another post below introduces how to depoloy your web app on an Apache web server. Should you have a Nginx web server, refer to the official guide.

Host Flet web app behind Apache web server by Reverse Proxy

Build as a web app

Complete the preparation steps 1 through 12 in the other post. Build options won’t make much difference for web, so simply run the command below. It takes some time to complete.

flet build web

Test locally

Built files are placed in build/web. Let’s test locally before pushing to a server. Execute the command and open the URL (http://localhost:8000) in your web browser.

python -m http.server --directory build/web
# Press Ctrl + C to exit.
On Chrome it works as expected.

Little more steps to upload

Specify the directory name

In this example, the web app will be deployed to https://blog.peddals.com/fletpassgen, so change the path in index.html. (By adding a build option --base-dir "/directoryname/" you can avoid this step, but you cannot test locally.) Edit index.html like the below. Make sure you have the directory name between slashes (/).

  <base href="/fletpassgen/">

Compress the entire folder

Change the folder name from web to the directory name, and compress it as a single file. You get fletpassgen.tar.gz as a result of these commands:

cd build
mv web fletpassgen
tar cvzf fletpassgen.tar.gz fletpassgen

Upload and extract

Upload the compressed file

To upload the compressed file to a hosting server, this example uses the scp command in Terminal.app. Replace username, hostname and upload directory based on your account details.

scp fletpassgen.tar.gz username@hostname:~/public_html

Login server and extract the file

If ssh is allowed, login your server and extract the file like the below. The directory has to be extracted in the correct location. In this example the web app will be in the subfolder /fletpassgen/ so it’s extracted in the document root of the website.

ssh username@hostname
cd ~/public_html
tar xvf fletpassgen.tar.gz
rm fletpassgen.tar.gz

Access the site by web browser

Your web app is ready now. URL should be like this: https://blog.peddals.com/fletpassgen/

After showing an icon for some moments then your web app will start working.

Few things to check if it’s not working

With this building method (static web app), the total size of files tends to be big. My example resulted 28MB in total. As the first access will take some time to download all required files, you have to be patient and wait until the app to be ready to start.

In case you don’t even see the icon after several seconds, take a look into the directory name in index.html, actual name of the extracted directory, user/group ownership and access permissions.

Tips and notes

Use same code for desktop and web apps

You may notice the layout of contents is broken when opening your app in web browser (I did!) Use ft.Container to place contents and width= property having the same value as page.window_width= so the horizontal layout will be kept in a wider window. For a simple app, having the below parameters keeps your app at the top center even in a web browser.

    page.vertical_alignment = ft.MainAxisAlignment.START
    page.horizontal_alignment = ft.CrossAxisAlignment.CENTER

File size is big

As I wrote, even this small app (Python code is approx. 3.9KB) becomes 28MB in total after a build. You need to put your eye on the available disk space.

App keeps running once loaded

Since this deployment method does not require a code to be running on the web server, your app keeps running in a web browser even when the network is down. For a simple tool it can be an advantage (I don’t know who needs a new password when offline, though).

Copy button won’t work on Safari (macOS and iOS)

This is a known issue. Hopefully it will be resolved in the near future, but at this moment copy works on Chrome but not on Safari. I added a code to hide the copy button based on the user agent, but it does not work. flet build web deployment won’t be able to get user agent unfortunately. You need to deploy as a server-side app if you want to add browser specific features.

Bonus: add Google AdSense advertisement

You can find this type of information for Flutter quite easily, but some of the ways I found didn’t work for my Flet app. If you’re looking for a solution, try this:

Get AdSense strings

Login your Google AdSense account, create new ad or click on the Get code < > icon of an existing ad to get strings.

Google AdSense > Ads > By ad unit > Display ads > Give it a name and Create > take note of the below two lines.


Add style to index.html

Edit the index.html file in the Flet web app directory and add the below CSS code, right above the </style> tag. Line numbers are just reference (Flet ver. 0.19.0).

        width: 100%;
        height: 100px;
        text-align: center;
        padding: 0;
        position: absolute;
        bottom: 0;
        z-index: 100;

Add <footer></footer> block to index.html

Right above the last line </body></html> of the index.html, add the below code. Replace highlighted xxxxxxxx and yyyyyyyy with strings you copied.

    .example_responsive_1 { width: 320px; height: 100px; }
    @media(min-width: 500px) { .example_responsive_1 { width: 468px; height: 60px; } }
    @media(min-width: 800px) { .example_responsive_1 { width: 728px; height: 90px; } } 
        <script async src="https://pagead2.googlesyndication.com/pagead/js/adsbygoogle.js"></script>
        <!-- Home_Page -->
        <ins class="adsbygoogle"
    (adsbygoogle = window.adsbygoogle || []).push({});

Save the file. You’ll see a horizontal ad at the bottom of the web page. If you just created a new ad, it may take some time for it to appear. Enjoy!

Image by Stable Diffusion

2024-Jan-29 0:04:44


512 x 512

Include in Image:
masterpiece, best quality, retro future, successful upload of application

Exclude from Image:



Guidance Scale:


ML Compute Unit:
CPU & Neural Engine

How to build Python GUI app on macOS with Flet

Flet version 1.18.0 released on the last day of year 2023 introduced a newer way of building apps that run on other computer, smartphone or website without setting up Python and Flet environment. I tried to build a macOS app and encountered a weird issue — a plain window opened with a text “Hello, Flet!” and I could not figure out how to fix it. A few days later, newer version 1.19.0 was released, and it had multiple fixes for build related issues. In this post I’d like to share how I built macOS app in Python with Flet. Once you set up your development environment, building process is really straight forward.

Official announcement in discord: https://discord.com/channels/981374556059086931/981375816489402408

Official GitHub change log: https://github.com/flet-dev/flet/blob/main/CHANGELOG.md

Official flet build guide: https://flet.dev/docs/guides/python/packaging-app-for-distribution

Quick introduction of my sample code

As an example I’m going to use my simple Password Generator app. The main purpose of this post is to introduce how to build GUI app with Flet, so I’m not going to write about too much of this sample code. As you can see in the screenshot below, you can edit number of characters and special characters to use. Copy button copies the password to clipboard. It’s fun to see how strong passwords the app generates by checking with online services like this one (bitwarden).

If you’re interested you can grab the code from this GitHub: https://github.com/tokyohandsome/passgen.py/blob/main/fletpassgen.py

Sample app you can build by following this guide. Number and special character set are editable.


  • macOS: 14.2.1
  • Python: 3.11.6
  • Python virtual environment: pipenv (version: 2023.10.24)
  • Modules: flet (version: 0.19.0)
  • Rosetta
  • Xcode: 15.2
  • Git: 2.29.2
  • cocoapods: 1.14.3_1
  • Flutter: 3.16.7

High level steps

  1. Build a virtual environment.
  2. Install Flet (version 0.18.0 or newer. 0.19.0 used in this post)
  3. Write a Flet app code or grab one.
  4. Test the code in your environment.
  5. Install requirements such as Flutter to build Flet app.
  6. Create a folder and work in there.
  7. Create assets folder and place an icon image.
  8. Copy your Flet code as main.py.
  9. Change the last part of main.py to ft.app(main).
  10. Create requirements.txt and add required modules.
  11. Clone build template from official GitHub.
  12. Edit copyright in the template.
  13. Build your app.

Little more detailed initial steps (#1~4)

Use your favorite virtual environment tool. Python version has to be 3.8 or newer. Install flet by executing pip install flet without specifying the version or flet==1.19.0. If you have not written your Flet app code yet, grab one from somewhere like the official site or from my GitHub if you’d like. In this post I’m going to use the fletpassgen.py as an example. Confirm it works by python3 fletpassgen.py then move forward.

Install requirements such as Flutter to build Flet app. (step #5)

To build as a desktop/smartphone/web app, you need to install Flutter, Dart, and a few other stuff based on the requirements. Once this process is done, you don’t need to redo unless anything goes wrong. Below steps are for Apple Silicon macs specifically. Skip any step if you’re already done.

1. Rosetta: Execute below to install.

sudo softwareupdate --install-rosetta --agree-to-license

2. Xcode 15: Download Xcode 15 from Apple’s website. Click on Download at the top of the page then double click on the installer.

3. Cocoapods: Execute below to install.

brew install cocoapods

4. Git: Execute below to install.

brew install git

5. Flutter: Follow the steps at the middle of the page and download Flutter SDK for your CPU (Intel or Apple Silicon such as M1, M2, M3…). Move the zip file to somewhere like ~/development/, unzip (double click), and add path to the commands to your PATH variable (rename handsome to your username, add below line to your ~/.zshrc then execute source ~/.zshrc to load the setting).

export PATH="/Users/handsome/development/Flutter/flutter/bin:$PATH"

Preparation for build (steps #6~12)

When you’re done all the above steps, create an app folder, go inside and perform the remaining steps. In this example the app name is fletpassgen.

mkdir fletpassgen
cd fletpassgen
mkdir assets
open assets

The above last command opens the assets folder in Finder where you can place a 512×512 pixel icon image named like icon.png (supported formats include .bmp, .jpg, and .webp). I don’t have anything else, but you can place other resources such as audio and text files used by our code.

Side note about icon: Flet can build app without an icon provided - it uses its own icon. I used Mochi Diffusion, a desktop app version of the AI image generator Stable Diffusion to generate the app icon which is also the top of this post. Prompts, model, etc. can be found at the last part of this post.

Next, copy your Flet app code as main.py.

cp ../fletpassgen.py main.py

If the last line of the code to call the main function is not ft.app(main), change so.

#if __name__ == "__main__":ft.app(target=main) <-- this needs to be changed to the below:

requirements.txt has to have needed Python module to run your Flet app, but pip freeze > requirements.txt caused several errors for me. In this example, flet was only needed (other imported modules are Python built-in). See the official guide for details especially when you’re building iOS or Android app.


You can clone the build template from the official GitHub to your local folder and edit copyright which can be shown by Command + I (macOS). Use your favorite editor if you are not good at the vi (vim) editor.

git clone https://github.com/flet-dev/flet-build-template
vi flet-build-template/cookiecutter.json

I edited highlighted lines 7-9 of cookiecutter.json in the flet-build-template folder.

    "out_dir": "",
    "python_module_name": "main",
    "project_name": "",
    "project_description": "",
    "product_name": "{{ cookiecutter.project_name }}",
    "org_name": "com.peddals",
    "company_name": "Peddals.com",
    "copyright": "Copyright (c) 2024 Peddals.com",
    "sep": "/",
    "kotlin_dir": "{{ cookiecutter.org_name.replace('.', cookiecutter.sep) }}{{ cookiecutter.sep }}{{ cookiecutter.project_name }}{{ cookiecutter.sep }}",
    "hide_loading_animation": true,
    "team_id": "",
    "base_url": "/",
    "route_url_strategy": "path",
    "web_renderer": "canvaskit",
    "use_color_emoji": "false"

Build your app (final step #13)

If you don’t care of options, simply execute flet build macos and you get an executable app. Below longer command adds version number and the above edited copyright. (Correction on Jan 25, 2024, thanks to the issue answered by the author.) To specify the build template location, use --template option followed by a relative path.

flet build macos --build-version "1.0.1" --template flet-build-template

Give it some time to complete and when you see Success! congrats, your app is built! Under build/macos/ you can find your app fletpassgen.app. It took around 3 mins and 10 secs to build it on my M1 mac mini. Output of the successful build process log is like below:

Creating Flutter bootstrap project...OK
Customizing app icons and splash images...OK
Generating app icons...OK
Packaging Python app...OK
Building macOS bundle...OK
Copying build to build/macos directory...OK
Screenshot in color

App is Universal

Type: Application (Universal). See version 1.0.1 and Copyright (by Cmd + I) as well.

Just like other mac apps, you can move the app to your Applications folder and double-click to launch. I was not asked to approve in the Privacy and Security. Not tested yet, but it should run on an Intel mac as well since it’s a Universal app. It opens and works just like the Python code. My sample app opens the Flet default size window momentarily then resizes as specified – I suppose there’s a better way to code.

Do I like it?

I love it! I used to use tkinter and pysimplegui to build desktop apps, but Flet is much easier with better looking/modern interface. Building process is straight forward, and app works great. Just like 3D printer changed productivity of nonprofessional DIYers in the real world, Flet lets you make your ideas real on your computer and share with your family, friends, colleagues and others. Hope you find it useful and enjoy as well!

Image by Stable Diffusion

2024-Jan-15 23:05:04


512 x 512

Include in Image:
masterpiece, best quality, retro future, cyber, disco computer, password generator

Exclude from Image:



Guidance Scale:


ML Compute Unit:
CPU & Neural Engine

(solved) macOS Finder didn’t open Documents and subfolders directly.

Not sure when it started, but Finder didn’t open ~/Documents and its subfolders directly. Somehow I managed to get it fixed. If you have the same issue, it can be fixed.

Decided to troubleshoot (finally).

Current running macOS version is Sonoma 14.2.1. I might have first noticed this issue when the macOS version was Big Sur 11.0 — Finder didn’t open folder by executing open ~/Documents/Python/fuga in Terminal.app, but it was not a big deal at that time. Today, I was playing with GitHub Desktop app and once again noticed the issue — the Show in Finder button didn’t open the local repo but my home directory instead. Something was (still) wrong.

↑ this button opened the home directory ↓ instead of the local repo folder (my UI is Japanese.)
Local repo located a couple folders under the Documents folder


Open command in Terminal didn’t open any folder under ~/Documents in Finder. I was able to navigate to lower levels folders in the same Finder window, so I had the right permissions. Subfolders in ~/Music, ~/Downloads, etc. had no problem to be opend. Go menu > Recent Folders and right-click > Open in New Tab resulted the exact same issue. Only Documents and its subfolders didn’t open directly in Finder in any way.

How I fixed it.

I tried multiple things for a couple hours and nothing helped. Finally, changing the folder view to List from Column seemed to have worked. Little more details follow: Open the Documents folder in Finder then select List from the View icon or drop-down.

or List from drop-down

From the window’s action menu or Finder’s View menu, select Show View Options to open the settings window. Command + J shortcut does the same thing.

Check “Always open in list view” then click on “Use as Defaults” at the bottom and close the window. Now the Documents and its subfolders open in the list view directly by the open command in Terminal.

I won.

Finder still changes view of window somehow whether you like it or not. So, I moved around/opened new windows/changed views/closed windows several times to reproduce the issue, but it never happened again (so far). I don’t know if the above steps was the right way, but the issue is gone.

Things that I tried but didn’t work/help, just for your info.

  • Search in the internet — could not find the exact same issue.
  • Reboot
  • Boot in safe mode.
  • Run Disk First Aid
  • Remove all app’s Documents folder access permissions, reboot and granted again.
  • Compare permissions of Documents and other folders.
  • Check Console.app

Image by Stable Diffusion

2024-Jan-3 17:56:25


512 x 512

Include in Image:
comicbook-style, gray hair guy looking for a missing folder in a book library

Exclude from Image:



Guidance Scale:


ML Compute Unit:
CPU & Neural Engine

You may need to install tkinter separately if it’s gone.

Recently I started working on some small Python GUI programs. Last year on an Intel Mac I found tkinter was not good as it had some issues with Japanese input method so I gave up. It’s been more than a year and I have a M1 (arm) Mac so there might be some improvements with tkinter. Upgrading Python3 (and all programs) by brew was the first step I tried, then tkinter module was unable to be loaded any longer…

You always better read install log which gives you necessary info.

I didn’t really pay attention to the brew log when upgraded Python from 3.9.1 to 3.9.6 as it was successful and must be a minor update… Soon after the upgrade my Python scripts that import tkinter in a pipenv shell started giving me error “No module named '_tkinter'“. Lots of trials and errors referring to several web pages, tech threads, etc. didn’t help. Tried setting up pipenv from scratch, installed tcl-tk by brew, added PATH and some other tcl-tk variables to .zshrc to no avail. I finally decided to give up and went to bed – worst case scenario I need to uninstall and install all Python related things including brew… On the day-2, I was calmer than last night and started identifying the scope of the problem – ok, tkinter cannot be found even out of my pipenv. It’s really gone from macOS Big Sur at some point. Reviewed steps taken on the day-1 and it didn’t take much time to find the below:

% brew info python3
python@3.9: stable 3.9.6 (bottled)
Interpreted, interactive, object-oriented programming language
tkinter is no longer included with this formula, but it is available separately:
  brew install python-tk@3.9

Aha! If I read the brew install log once completed, I didn’t need to waste 6 hours and would have a sweet dream. Anyways, if your tkinter is gone after upgrading Python by brew, just execute another command brew install python-tk@3.9 and install it as well. My scripts finally open tkinter GUI windows with no error.

Most of Python documents or webpages suggest you to try tkinter for GUI development as it’s installed by default, but it’s not true anymore especially for those of you who install Python by brew on Mac.

Example of the error in my case

% python3 -m tkinter
Traceback (most recent call last):
File "/opt/homebrew/Cellar/python@3.9/3.9.6/Frameworks/Python.framework/Versions/3.9/lib/python3.9/runpy.py", line 188, in _run_module_as_main
mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
File "/opt/homebrew/Cellar/python@3.9/3.9.6/Frameworks/Python.framework/Versions/3.9/lib/python3.9/runpy.py", line 147, in _get_module_details
return _get_module_details(pkg_main_name, error)
File "/opt/homebrew/Cellar/python@3.9/3.9.6/Frameworks/Python.framework/Versions/3.9/lib/python3.9/runpy.py", line 111, in _get_module_details
File "/opt/homebrew/Cellar/python@3.9/3.9.6/Frameworks/Python.framework/Versions/3.9/lib/python3.9/tkinter/__init__.py", line 37, in
import _tkinter # If this fails your Python may not be configured for Tk
ModuleNotFoundError: No module named '_tkinter'

© Peddals.com