@cz_binance@GiggleAcademy@GigglesAcademy I am excited about the potential opportunity to work with YOU and contribute to the Giggle Academy. I would be delighted to discuss further how I can add value to your organization. I look forward to the possibility of working together @cz_binance
@cz_binance@GiggleAcademy@GigglesAcademy I am particularly interested in bringing my data storytelling expertise to the table. This approach goes beyond simple data visualization and focuses on providing context, insights, and actionable recommendations that can help drive business decisions.
Find @overnance on @EthereumDenver and take a selfie with himπΆοΈ
We're going to send you a special invitationποΈ to OverProtocol Open Beta Testnet Season 2.
Week 7 Notes - #buildinpublic
47 days in | At 32% of my Goal.
π΄π΄π΄π΄π΄π΄
Summary π:
Wow, it's already the seventh week! Although I didn't make as much progress as planned, I'm still optimistic because every win counts!
Booted up the build for a stress test for 3 minutes, and the CPU went above the acceptable temps, so I had to shut down. It's a no-brainer to get optimal cooling for me not to fry any components!
PS: I was happy it booted up π
To achieve optimal cooling, I opted for
πUpright riser for the second GPU
πLian Li Galahad 360 AIO Cooler
π3 sets of 3 Lian Li Fan ST120, ARGB 120mm Fan.
πLian Li Premium PCI-E 16X 4.0 Extender Riser Cable 600 mm
I used 10 fans, 4 as exhaust - 3 below the cooler radiator and 1 mounted at the rear case. The remaining 6 fans were used for intake - 3 behind the GPU and 3 below the case. I believe this will create positive airflow and keep everything cool!
I haven't bothered with cable management yet because I'm still waiting for the delivery of the Extender Riser Cable. I researched using the regular USB cable riser used in mining rigs.
However, from my findings, this is not recommended, hence my decision to opt for the standard riser cable.
As for the non-technical part, I spent time compiling assets to train the LoRa Model. I now have over 200 images.
Since most of the assets I needed were scarce, I became resourceful. I started hunting for documentary videos, taking high-quality screenshots using VLC.
Will this work?
They look high-res; I will share my results when I reach that stage.
π΄π΄π΄π΄π΄π΄
The Goal π:
βUse AI to generate art depicting the Yorubas' culture from the 17th Century to the present in a time series or era-based format.
π΄π΄π΄π΄π΄π΄
The Strategy π§ :
βUnderstand how GANs and Ultra Realistic AI Arts work.
βDiscover Yoruba artists representing various styles and contributions to Yoruba art, including the Lucumi from Cuba and Brazil.
βCategorize the time period into four sections, namely pre-19th Century, 19th Century, 20th Century, and 21st Century
βOnce I have a collection of creatives, the next step is to develop a portfolio website to showcase and receive feedback.
π΄π΄π΄π΄π΄π΄
The plan π:
βStage 1 - Research and Discovery: 25% Done
βStage 2 - Consume and Activate: 12% Ongoing
βStage 3 - Showcase Results
βStage 4 - Feedback and Share Lessons Learned
π΄π΄π΄π΄π΄π΄
Next Steps π:
π From Last week:
β³ Finish building the hardware β
β³ Install all required software and pull all repo files β
β³ Learn how to train LoRa models using Kohya_ss GU (this may take time) β
β³ Collect datasets looking at 30 - 100 assets β
π Moving Forward:
β³ Finish building the hardware
β³ Install all required software and pull all repo files
β³ Learn how to train LoRa models using Kohya_ss GU (this may take time)
π΄π΄π΄π΄π΄π΄
What I've Learned π€:
β Airflow configuration and AIO cooling
β Sourcing images from videos, sneaky!
β Believe me, prioritization on a God mode!
β Risk management ++
________
The Challenges π’:
β Managing commitments: last week was a bit overwhelming!
β As usual, my prompts, but I am getting closer each day!
π΄π΄π΄π΄π΄π΄
Total costs So Far π΅: I was advised not to include these parts anymore. But my DM is open if you want to know the details.
π΄π΄π΄π΄π΄π΄
Revenue So Far π°: $0 - Non-profit venture
π΄π΄π΄π΄π΄π΄
THANK YOU if you have made it this far! What a brief summary! π
If you have any questions or suggestions for me, please feel free to leave them in the comments or send them to DMs.
Thank you and GN. ππ»
My image here is that of the Eyo Masquerade!
Also known as Adamu Orisa, he is the main character in a prominent cultural and traditional festival of the Yoruba people, in Lagos.
Week 6 Notes - #buildinpublic
40 days in | At 35% of my Goal.
π΄π΄π΄π΄π΄π΄
Summary π:
In week 6, I had to make a major decision regarding my PC build. My initial plan was to use dual GPUs, but after consulting with BestBuy's Geek Squad, I was convinced to use only one GPU.
They explained that having a second GPU would be unnecessary, as my PC would only utilize it partially. Additionally, housing two RTX-3090s requires some serious cooling strategies. So, I changed my plan and got all the necessary components for a single GPU build.
But the next day, Nvidia released a new feature that allows applications to share VRAM - talk about perfect timing!
It's called System Memory Fallback for Stable Diffusion. You can find more details here: https://t.co/uWm8rWA1N1.
So, once again, another pivot in my plan! As they say, if something is worth doing at all, it's worth doing well.
Let me walk you through my updated configuration:
- GPU: RTX 3090 2x
- CPU: Intel Core i7-12700K
- MOBO: GIGABYTE Z690 AORUS ELITE AX
- RAM: DDR4 48GB
- SSD: 1tb Nvme SSD
- Case: Dynamic EVO XL - Full Tower (the largest case I've ever seen!)
- Cooler: Noctua NH-L9i
- PSU: EVGA 1600w fully modular
For more detailed information on each component, check out this link on PC Part Picker: https://t.co/Kmthr2bLNx
Now, putting aside the technical aspects for a moment...
In preparing for the machine learning tasks, I started collecting and compiling training assets, and I couldn't help but be amazed by the incredible craftsmanship and advancements achieved then as well as today within this culture.
Unfortunately, though, this rich culture and its artifacts have not received the attention they deserve, with many stolen and locked away in European museums.
π΄π΄π΄π΄π΄π΄
The Goal π:
βUse AI to generate art depicting the Yorubas' culture from the 17th Century to the present in a time series or era-based format.
π΄π΄π΄π΄π΄π΄
The Strategy π§ :
βUnderstand how GANs and Ultra Realistic AI Arts work.
βDiscover Yoruba artists representing various styles and contributions to Yoruba art, including the Lucumi from Cuba and Brazil.
βCategorize the time period into four sections, namely pre-19th Century, 19th Century, 20th Century, and 21st Century
βOnce I have a collection of creatives, the next step is to develop a portfolio website to showcase and receive feedback.
π΄π΄π΄π΄π΄π΄
The plan π:
βStage 1 - Research and Discovery: 25% Done
βStage 2 - Consume and Activate: 10% Ongoing
βStage 3 - Showcase Results
βStage 4 - Feedback and Share Lessons Learned
π΄π΄π΄π΄π΄π΄
Next Steps π:
π From Last week:
β³ Get the missing PC parts β and finish building the hardware β
β³ Install all required software and pull all repo files β
β³ Learn how to train LoRa models using Kohya_ss GU (this may take time) β
β³ Collect datasets looking at 30 - 100 assets β
π Moving Forward:
β³ Finish building the hardware
β³ Install all required software and pull all repo files
β³ Learn how to train LoRa models using Kohya_ss GU (this may take Time)
β³ Collect datasets looking at 30 - 100 assets - Ongoing
π΄π΄π΄π΄π΄π΄
What I've Learned π€:
β Hardware setups, SLI, and airflow!
β You can't capture the real essence of a specific culture if the generative model hasn't been trained with information on them. I tried multiple prompts for unique artifacts peculiar to the Yorubas but failed to generate tangible outputs.
β The richness of Yoruba culture through legacy arts proves a high level of civilization pre-colonialization.
β I was shocked to discover attempts by non-yoruba's telling our stories!
________
The Challenges π’:
β As usual, my prompts, but it's a WIP
β Time is still the same issue, but I am improving a day at a time.
π΄π΄π΄π΄π΄π΄
Total costs So Far π΅: I was advised not to include these parts anymore. But my DM is open if you want to know the details.
π΄π΄π΄π΄π΄π΄
Revenue So Far π°: $0 - Non-profit venture
π΄π΄π΄π΄π΄π΄
THANK YOU if you have made it this far! My summaries are always a long read! π
If you have any questions or suggestions for me, feel free to leave them in the comments or send them via direct message.
Thank you. ππ»
My prompt: neon cubist courtesy @HBCoop_
Week 5 Notes - #buildinpublic
33 days in | At 30% of my Goal.
π΄π΄π΄π΄π΄π΄
Summary π:
I had planned to complete my research and discovery phase by examining Yoruba artwork in the Modern Era from the 20th Century to the present during the fifth week.
However, I got sucked into AI and machine learning, much like Alice falling into a rabbit hole! But just like McDonald's says, 'I'm lovin it!' πβοΈ
But first, let me recap the non-techy. I explored contemporary manifestations of Yoruba culture through various forms of art that showcased modern-day Yoruba life.
This included documenting Adire clothing and modern fashion trends as well as exploring contemporary Yoruba music, entertainment industry influences, and finally, looking at how urbanization has impacted their cultural heritage.
On the technical side of things, I am pleased with my progress so far. The prompts I used gave the desired results after a few attempts.
I used GPT4 to optimize them before running them through MidJourney and Dalle-3. I even tried experimenting with Seeding, which was quite interesting.
Out of curiosity, I played around with image to video using Pika; after mistakes and consultations with Ben @bennash & Jerrod @OnwardsProject (special thanks to both of you), I understood that the more objects within an image, the trickier it became to control morphing effects.
This was when my journey truly began down the "rabbit hole!" I tried several prompts to explore wood sculpturing in the culture's style but failed, and I got frustrated.
So, went back to the drawing board and discovered I missed a critical step, "understanding how GANs work," which I highlighted in my initial findings.
Long story short, I discovered Stable Diffusion, an open-source solution. I was surprised that I can do text-to-image, image-to-video, and text-to-voice by training a model with my own style, on my own HARDWARE, with no limitations, and above all, 100% privacy! π²
So, I will be suspending all subscriptions except for OpenAI. MJ and Dalle-3 are great tools, but the next step is to mix Synthetic (AI) Data with "real" data into training to get that unique output with a piece of me weaved into it.
Moving forward, I will use stable diffusion XL with LoRas trained using my data.
If I ever need it, I will use Animatediff for motion, Amazon Poly for voice cloning, and Adobe Express to mix stuffs up whenever needed.
For my hardware, I have a set of RTX GPUs I used for mining ETH for a year or less (2x rtx3090 & 2x rtx3070) with an old B365 HD3 2-way crossfire compatible MOBO and a brand new Intel Core i5-9600K Coffee Lake CPU.
What is missing are PSU, RAM, and SSD.
Potential Hardware Config will look like:
GPU: RTX 3090 2x
CPU: Intel Core i5-9600K Coffee Lake
MOBO: Gigabyte B365 HD3
PSU: EVGA 1300w fully modular plus gold or Plat.
RAM: DDR4 min 32gb & max 64gb. I need VRAM
SSD: 1tb or 2tb Nviem SSD
Case: Airflow Super Tower - eyeing Corsair 4000Ds
Cooler: Not yet decided - Water cooling, I guess.
Although, folks claim SDXL can't run multiple GPUs. I will try a repo I saw: Multi-threaded GUI manager for mass creation of AI-generated art with support for multiple GPUs.
π΄π΄π΄π΄π΄π΄
The Goal π:
βUse AI to generate art depicting the Yorubas' culture from the 17th Century to the present in a time series or era-based format.
π΄π΄π΄π΄π΄π΄
The Strategy π§ :
βUnderstand how GANs and Ultra Realistic AI Arts work.
βDiscover Yoruba artists representing various styles and contributions to Yoruba art, including the Lucumi from Cuba and Brazil.
βCategorize the time period into four sections, namely pre-19th Century, 19th Century, 20th Century, and 21st Century
βOnce I have a collection of creatives, the next step is to develop a portfolio website to showcase and receive feedback.
π΄π΄π΄π΄π΄π΄
The plan π:
βStage 1 - Research and Discovery: 25% Done
βStage 2 - Consume and Activate: 5% Ongoing
βStage 3 - Showcase Results
βStage 4 - Feedback and Share Lessons Learned
π΄π΄π΄π΄π΄π΄
Next Steps π:
π From Last week:
β³ Tried more prompts with MidJ and Dalle-3 prompts β
β³ Failed with fine-tuning output to my desired level β
β³ Explored modern-day manifestations of Yoruba Culture β
π Moving Forward:
β³ Get the missing PC parts and finish building the hardware
β³ Install all required software and pull all repo files
β³ Learn how to train LoRa models using Kohya_ss GU (this may take time)
β³ Collect datasets looking at 30 - 100 assets
π΄π΄π΄π΄π΄π΄
What I've Learned π€:
β AI tech is open-source, it's very powerful, and it's here to stay!
β Understood AI Workflows using comfyui
β You can make 20 - 30-minute AI clips, π€£
________
The Challenges π’:
β My prompts, but I think this will be resolved soon.
β Balancing my research with other commitments is still tricky, but scheduling my tweets and automating some of my grunt tasks is helping.
π΄π΄π΄π΄π΄π΄
Total costs So Far π΅: I was advised not to include these parts anymore. But my DM is open if you want to know the details.
π΄π΄π΄π΄π΄π΄
Revenue So Far π°: $0 - Non-profit venture
π΄π΄π΄π΄π΄π΄
Once again, if you have made it this far, THANK YOU!
It's a long read!
I'd love to hear your thoughts on how I can improve my experience, and if you have any questions, please leave them in the comments or send them via DM.
Thank you.
My prompt: Transhumanistic
Pika labs rocks