Release V1 Published on GIthub

I had to rewrite a lot of the code and test out different versions of dependencies to make the entire project fit under a 2GB .exe file.

Sadly, due to the extremely huge size of the drivers, the released version is about 2.6gb in size. I can reduce it to under 500mb, but I found out that downloading the dependencies during the First Boot would take more time than to just download the 2.6gb project with most dependencies preinstalled.

Also, I did not include the AI models in the release files as it is 12+GB, and upon the first boot it would be downloaded and set up automatically, which may take 5-10 minutes.

Challenges

I was having an extremely hard time with Pytorch and stuff because it compiled a lot of useless junk which made the exe file extremely big. I had to manually check which dlls were being used and which not, and had to remove the useless ones. Also, for some reason it just refused to detect my GPU after building to the .exe, which I fixed afterwards.

Also, the method for reading files which I was using with os would not work in an exe form, as the files get unpacked into a temp folder somewhere in appdata. I had to get that sorted too.

I have now released the proper requirements file, along with the working built exe and all the files on github.

Image: I also made a good looking banner for the project.

If you are reading my devlogs for voting. Please start from the bottom, from the very first devlog.

All Set, READY FOR FIRST SHIP!

After all this work, The Clueless Closet has now reached its first stable release and is ready for its first ship.

Changelog

• Cleaned up code
• Added better error handling
• Added CPU ONLY mode for unsupported GPUs

Features

• Use of AI to detect type of Clothes of User
• Use of AI to create text from the data
  (All the AI Processing is done LOCALLY! No API Keys needed)
• Weather Data collection and processing
• GPU Based AI & Multithreading Enhancements
• Editing images through Python (Snapshot Feature)
• Use of tkinter to make Menus
• Personalities of AI
  And more..

Future Updates:

On the V2, I would be working on making the snapshots a lot better.
Snapshots would first use AI to remove background from your image, then put your image on a template, add all information and the AI’s response on the template and make a perfect high quality photo.

More features coming in v2!

Attached Video: Shows the program running, My face and camera feed is blacked out (for privacy reasons)

The terminal panels you see in the video are from my Pytterns python framework which is also a hack club project. Can be viewed through the following links:

https://flavortown.hackclub.com/projects/14732
https://github.com/brahmtej2009/The-Clueless-Closet/

Changelog

• Used AI to make a PERFECT UI
• Added Dad Mode (for Dad Jokes)
• Added Hypeman Mode backgrounds
• Made weather conditions logic more accurate
• Emojis in UI
• Fixed bug where buttons would merge colors when clicked.

Challenges

I had to do graphic designing and stuff to make the backgrounds a repeatable tiles which was a bit tricky to get 100% right, but I did it through CANVA.

Learning & Overcoming Challenges

• I learnt how to make better Tkinter UI
• I learnt how to make repeating tiles
• I learnt how to make Dad Jokes
• I learnt that gemini is trash at Tkinter

Future Updates

I would be adding features, which would take user’s image, remove background, add it to a template background, and nicely add the roast text in that template for the snapshot. Which would be different for every personality mode. GREAT! (Coming Soon!)

AI Disclosure

After using gemini for my UI Help, which basically sucked, I tried perplexity. It gave me good ideas for my UI, like the emojis. Parts of the UI are coded by the AI. All the codeblocks which had AI Involvement are marked through comments in the code with full disclosure.

Image: Attached image shows the new UI in Dad Mode (Dadjokes mode)

Thanks for reading this 💖

BIG UPDATES!

In the past 3 hours, I have almost ENTIRELY chanted the UI, and added Great Features!

Changelog

• Every personality has its own UI
• Made it easier to add new personalities for anyone
• Personality Based UI and Colors
• Snapshot Backgrounds based on Personality
• Improved Layout and Interface
• More efficient AI and better performance

Challenges

I had to do graphic designing and stuff to make the backgrounds a repeatable tiles which was a bit tricky to get 100% right, but I did it through CANVA.

Learning & Overcoming Challenges

• I learnt how to use Canva to make Repeating Tiles
• I learnt how to make better tkinter interfaces
• I learnt how to make AI better and faster through better prompting

Future Updates

I would be adding features, which would take user’s image, remove background, add it to a template background, and nicely add the roast text in that template for the snapshot. Which would be different for every personality mode. GREAT!

AI Disclosure

Gemini was used to help with tkinter menus and helped in background images. All the codeblocks which had AI Involvement are marked through comments in the code with full disclosure.

Thanks for reading this 💖

See the attached image, its soooo cute dammnn

Changelog

Personalities:

• Grandma Personality

An always loving and ‘Awwww my baby’ personality

• Hype Man

An always ‘cheering up’ personality

• Roast

Rude and shaming personality

Challenges

I have made the system for the personalities. Linked it with the settings. But prompting the AI to give the PERFECT responses is timetaking and hard. Each time I edit, I have to reboot the project, which takes 2-4 minutes. Also, for lower pc hardware systems, personalities dont seem to work that good due to hardware limitation.

Learning & Overcoming Challenges

I learnt how to give personalities to AI through prompting

Future Updates

I am thinking to add different backgrounds in snapshots based on moods.

Also, looking forward to completely remake the tkinter interface for every personality.

I am thinking about maybe also making a character which appears on the screen when the program is loaded, showing expressions. (Maybe…)

AI Disclosure

No significant use of AI was done for this devlog. All the codeblocks which had AI Involvement are marked through comments in the code with full disclosure.

Image: The Gramma Mode giving PERFECT REPLIES!! AWW this is soo cute <3

Thanks for reading this 💖
Stay Tuned!

Changelog

• Personalities (Work in Progress)

Challenges

It’s been over an hour. I am working on prompts and ideas for all types of personalities I can make. I have perfected the Roast Personality. Trying to make other personalities work as good as it.

Learning & Overcoming Challenges

I learnt how to construct prompts in code through different factors

Future Updates

I am thinking to add different backgrounds in snapshots based on moods.

Also, looking forward to completely remake the tkinter interface for every personality.

AI Disclosure

No significant use of AI was done for this devlog. All the codeblocks which had AI Involvement are marked through comments in the code with full disclosure.

Image: All the settings parameters I am working on right now

Thanks for reading this 💖
Stay Tuned!

Changelog

• High Performance Multi Threading
• UI Buttons Updates
• Changed AI Model for Roast
• Hardware based Responses

Challenges

The amount of critical roast I was looking for needed a Bigger AI model, and more processing time, which caused timeout issue on tkinter. Multithreading was added to keep it stable. With this, the buttons update according to the process.

For weaker systems, this heavy AI would not work, so it would automatically fall back to the easier to run AI, at the cost of roast quality.

Learning & Overcoming Challenges

I learnt how to add popups in tkinter
I learnt how to run functions in threading
I learnt how to get GPU & Process information

Future Updates

I would be working on different personalities. I feel this has much more potential than just a roasting AI. It could literally be a Gramma or even Groot (I am Groot!)

AI Disclosure

Slight use of Gemini was done for bugfixing for multithreading. All the codeblocks which had AI Involvement are marked through comments in the code with full disclosure.

Image 1: Attached image shows the AI scanning the image in threaded mode while tkinter shows a proper interface of it.

Image2: Attached image shows the Final output after scan is done. The snapshot button got activated.

Image area is black for privacy reasons.

Thanks for reading this 💖
Stay Tuned!

Changelog

• Diary Logging
• Snapshot Button
• Better error handling & Cleaned up code

Challenges

For this update, I had to take the live image from feed, use python to add a white bar on top of the image, and write text and save it. This was a BIG thing for me, I managed to do it through cv2 ultimately.

Learning & Overcoming Challenges

I learnt how to edit images through code
I learnt how to add text to images
I learnt how to use .tiff (Text Font) Files

Future Updates

I’ll also try to auto crop the image somehow so it only contains the user.

I would also be working on improving the AI Model. Making 2 separate models. One for high power systems (Gives good result) and one for low power systems where I would use a smaller AI Model.

AI Disclosure

Not much AI was used in this update. All the codeblocks which had AI Involvement are marked through comments in the code with full disclosure.

Image: Attached image shows the rendered image which comes from the Snapshot (And gets saved)

Image area is black for privacy reasons.

Thanks for reading this 💖
Stay Tuned!

Changelog

• Complete UI Rewamp
• Addition of Weather Stats in UI
• Better display of roasts in UI.
• Feature to change the camera for systems with multiple cameras

Challenges

I designed the old tkinter UI myself, which was quite basic and unappealing. Making a new looking UI, with all the things displaying properly without looking crowded was a good challenge. I had to take some help from Gemini in order to design the new UI, and now it looks sooo much better with the colors and all.

Learning & Overcoming Challenges

I learnt how to make better tkinter menu
I learnt how to add stuff to UI using Grid Method and align it properly

Future Updates

As the UI is now new, I would be working on more features, and more buttons. I’ll also be adding more UI designs or something in settings which user can change according to their liking.

I’ll also try to auto crop the image somehow so it only contains the user.

Il also try to make a daily diary which gets generated automatically inside a folder, with a history of roasts on the image. (Next Update)

AI Disclosure

I used Gemini to help me make the Tkinter UI, It explained me properly how to use grid and pack properly to align my stuff. All the codeblocks which had AI Involvement are marked through comments in the code with full disclosure.

Image: Attached image shows the NEW UI (IMG 1) and the Old UI (IMG 2)

Thanks for reading this 💖
Stay Tuned!

Changelog

• Great performance enhancements
• GPU AI Generation on system with Nvidia GPUs
• Switching from GPU / CPU based processing
• Added requrements.txt with proper versions of all libraries

Challenges

I have never worked with GPUs in the past, for this project, i had to utilize stuff like torch.cuda and had to integrate that with my AI models. AI tools helped me a lot in this to get a basic understanding of how all this worked. I also was unable to reduce the startup times and processing time of the AI. Which I would be working on soon.

Learning & Overcoming Challenges

I learnt how to use GPU in AI Programs
I learnt how to Improve Code Performance
I learnt how to Detect Presence of Nvidia/Apple Silicon in a system

Future Updates

The tkinter interface is really old looking, I would be working on that in the future release. I’l try to display the information like temp and rain etc. on the screen along with the live feed.

Il also try to make a daily diary which gets generated automatically inside a folder, with a history of roasts on the image. (Most likely Next to Next devlog)

AI Disclosure

I used Gemini in this for research, as I am quite a newbie in this. I got a lot of ideas while conversation with AI. I used AI for learning use of GPUs in AI Projects. All the codeblocks which had AI Involvement are marked through comments in the code with full disclosure.

Image: Attached image shows the system detecting the GPU and automatically switching to it to increase system performance. This output would most likely not be shown on the tkinter frontend as its not required.

Thanks for reading this 💖
Stay Tuned!

Changelog

• Changed from Static Roasting to Live Generated Roast based on all Factors
• Performance enhancements (Quicker Startup)
• Better AI Detection
• Roast is now displayed on the software panel, under your live feed
• Completely removed the heat score logic
• Added another AI which dynamically takes information and generates LIVE ROASTS!

Challenges

As i have said before, I’m kinda a newbie in terms of AI, so this update was a bit timetaking for me. I also faced a lot of issues while trying to make the AI be rude and roasty, but at the same time, to avoid swearing. It took a lot of tries.

Learning & Overcoming Challenges

I learnt how to use text generative AI model
I learnt how to pass information properly to an AI Model
I learnt how to NOT do overfitting and let the responses be natural
I learnt how to mess with generation settings like temperature
I learnt how to teach AI to roast people 💀

Future Updates

I am aware that this project runs entirely on CPU right now, which is quite slow when it comes to AI Processing than GPUs. This causes lag in tkinter.

Il try to run the project on GPU & Minimize Lag

Il also try to make a daily diary which gets generated automatically inside a folder, with a history of roasts on the image.

AI Disclosure

I used Gemini in this for research, as I am quite a newbie in this. I got a lot of ideas while conversation with AI, like better prompting for roasting ai, picking up correct AI Models, Reducing processing time and more. All the codeblocks which had AI Involvement are marked through comments in the code.

Image: Attached image shows the AI roasting me for wearing a normal tshirt (While location was set to Canada, where temps are around -30 right now)

The image does not show the video feed as I am not comfy with revealing my face. :)

Thanks for reading this 💖
Stay Tuned!

I have completely changed the approach I used for AI. Instead of a normal classification model, I now use a Rank Based System where only most matching choice is selected.

Changelog

• Changed to patrickjohncyh/fashion-clip model
• Use of Rank Based Identification
• Weather fetching from openmeteo
• Integration of my own framework (Pytterns) for display of neat panels

Challenges

I am a complete newbie in terms of AI, I had to spend a good amount of time learning how to use Image Classification Models. I also had to search a lot and test a lot to find the correct model.

The camera also took a lot of time to load, and I found out that we had to use the DirectShow to make it faster for windows.

Learning & Overcoming Challenges

I learnt how to use openmeteo weather API
I learnt how to manipulate and make sense from all the weather factor data
I learnt how to use Classification AI Models
I learnt how to build live camera feed in python
I learnt how to display a live camera feed on tkinter properly

Future Updates

From data, the project must be able to identify mistakes in clothing, and provide good roasts
Make the heat score logic a bit better so it counts more factors
Make its logic so it combines all mistakes for 1 good solid roast

AI Disclosure

I used Gemini to get information about DirectShow & Live Feed Processing. All the work done by AI is properly marked in the code through comments. I always ensure AI Material is less than 40% of the project’s total code.

Image: Attached image shows the terminal output for calculation of Heat Score in the amazon basin. More information on how this is processed in future devlogs!

Thanks for reading this 💖
Stay Tuned!

The project is released on github.

It is functional, and currently provides information about clothes being Cool or Warm.

Integration with weather is still pending, which is the thing which I am working on.
I am also looking for some cool roasts, which I’l be adding to this by tomorrow’s update.

Looking forward to build a great fun thing!