Are you an idiot at fundraising?

Humans undervalue what they don't grasp. Consider NASCAR. How is that a sport? ask uneducated observers. Circular traffic. Driving near a car's physical limits is different from daily driving. When driving at 200 mph, seemingly simple things like changing gas weight or asphalt temperature might be life-or-death.

Venture investors do something similar in entrepreneurship. Most entrepreneurs don't realize how complex venture finance is.

In my early startup days, I didn't comprehend venture capital's intricacy. I thought VCs were rich folks looking for the next Mark Zuckerberg. I was meant to be a sleek, enthusiastic young entrepreneur who could razzle-dazzle investors.

Finally, one of the VCs I was trying to woo set me straight. He insulted me.

How I learned that I was approaching the wrong investor

I was constructing a consumer-facing, pre-revenue marketplace firm. I looked for investors in my old university's alumni database. My city had one. After some research, I learned he was a partner at a growth-stage, energy-focused VC company with billions under management.

Billions? I thought. Surely he can write a million-dollar cheque. He'd hardly notice.

I emailed the VC about our shared alumni status, explaining that I was building a startup in the area and wanted advice. When he agreed to meet the next week, I prepared my pitch deck.

First error.

The meeting seemed like a funding request. Imagine the awkwardness.

His assistant walked me to the firm's conference room and told me her boss was running late. While waiting, I prepared my pitch. I connected my computer to the projector, queued up my PowerPoint slides, and waited for the VC.

He didn't say hello or apologize when he entered a few minutes later. What are you doing?

Hi! I said, Confused but confident. Dinin Aaron. My startup's pitch.

Who? Suspicious, he replied. Your email says otherwise. You wanted help.

I said, "Isn't that a euphemism for contacting investors?" Fundraising I figured I should pitch you.

As he sat down, he smiled and said, "Put away your computer." You need to study venture capital.

Recognizing the business aspects of venture capital

The VC taught me venture capital in an hour. Young entrepreneur me needed this lesson. I assume you need it, so I'm sharing it.

Most people view venture money from an entrepreneur's perspective, he said. They envision a world where venture capital serves entrepreneurs and startups.

As my VC indicated, VCs perceive their work differently. Venture investors don't serve entrepreneurs. Instead, they run businesses. Their product doesn't look like most products. Instead, the VCs you're proposing have recognized an undervalued market segment. By investing in undervalued companies, they hope to profit. It's their investment thesis.

Your company doesn't fit my investment thesis, the venture capitalist told me. Your pitch won't beat my investing theory. I invest in multimillion-dollar clean energy companies. Asking me to invest in you is like ordering a breakfast burrito at a fancy steakhouse. They could, but why? They don't do that.

Yeah, I’m not a fine steak yet, I laughed, feeling like a fool for pitching a growth-stage VC used to looking at energy businesses with millions in revenues on my pre-revenue, consumer startup.

He stressed that it's not necessary. There are investors targeting your company. Not me. Find investors and pitch them.

Remember this when fundraising. Your investors aren't philanthropists who want to help entrepreneurs realize their company goals. Venture capital is a sophisticated investment strategy, and VC firm managers are industry experts. They're looking for companies that meet their investment criteria. As a young entrepreneur, I didn't grasp this, which is why I struggled to raise money. In retrospect, I probably seemed like an idiot. Hopefully, you won't after reading this.

Unexpected Success

Shopify CEO and co-founder Tobias Lutke. Shopify is worth $48 billion.

World-renowned entrepreneur Tobi

Tobi never expected his first online snowboard business to become a multimillion-dollar software corporation.

Tobi founded Shopify to establish a 20-person company.

The publicly traded corporation employs over 10,000 people.

Here's Tobi Lutke's incredible story.

Elon Musk tweeted his admiration for the Shopify creator.

30-October-2019.

Musk praised Shopify founder Tobi Lutke on Twitter.

Happened:

Explore this programmer's journey.

What difficulties did Tobi experience as a young child?

Germany raised Tobi.

Tobi's parents realized he was smart but had trouble learning as a toddler.

Tobi was learning disabled.

Tobi struggled with school tests.

Tobi's learning impairments were undiagnosed.

Tobi struggled to read as a dyslexic.

Tobi also found school boring.

Germany's curriculum didn't inspire Tobi's curiosity.

“The curriculum in Germany was taught like here are all the solutions you might find useful later in life, spending very little time talking about the problem…If I don’t understand the problem I’m trying to solve, it’s very hard for me to learn about a solution to a problem.”

Studying computer programming

After tenth grade, Tobi decided school wasn't for him and joined a German apprenticeship program.

This curriculum taught Tobi software engineering.

He was an apprentice in a small Siemens subsidiary team.

Tobi worked with rebellious Siemens employees.

Team members impressed Tobi.

Tobi joined the team for this reason.

Tobi was pleased to get paid to write programming all day.

His life could not have been better.

Devoted to snowboarding

Tobi loved snowboarding.

He drove 5 hours to ski at his folks' house.

His friends traveled to the US to snowboard when he was older.

However, the cheap dollar conversion rate led them to Canada.

2000.

Tobi originally decided to snowboard instead than ski.

Snowboarding captivated him in Canada.

On the trip to Canada, Tobi encounters his wife.

Tobi meets his wife Fiona McKean on his first Canadian ski trip.

They maintained in touch after the trip.

Fiona moved to Germany after graduating.

Tobi was a startup coder.

Fiona found work in Germany.

Her work included editing, writing, and academics.

“We lived together for 10 months and then she told me that she need to go back for the master's program.”

With Fiona, Tobi immigrated to Canada.

Fiona invites Tobi.

Tobi agreed to move to Canada.

Programming helped Tobi move in with his girlfriend.

Tobi was an excellent programmer, therefore what he did in Germany could be done anywhere.

He worked remotely for his German employer in Canada.

Tobi struggled with remote work.

Due to poor communication.

No slack, so he used email.

Programmers had trouble emailing.

Tobi's startup was developing a browser.

After the dot-com crash, individuals left that startup.

It ended.

Tobi didn't intend to work for any major corporations.

Tobi left his startup.

He believed he had important skills for any huge corporation.

He refused to join a huge corporation.

Because of Siemens.

Tobi learned to write professional code and about himself while working at Siemens in Germany.

Siemens culture was odd.

Employees were distrustful.

Siemens' rigorous dress code implies that the corporation doesn't trust employees' attire.

It wasn't Tobi's place.

“There was so much bad with it that it just felt wrong…20-year-old Tobi would not have a career there.”

Focused only on snowboarding

Tobi lived in Ottawa with his girlfriend.

Canada is frigid in winter.

Ottawa's winters last.

Almost half a year.

Tobi wanted to do something worthwhile now.

So he snowboarded.

Tobi began snowboarding seriously.

He sought every snowboarding knowledge.

He researched the greatest snowboarding gear first.

He created big spreadsheets for snowboard-making technologies.

Tobi grew interested in selling snowboards while researching.

He intended to sell snowboards online.

He had no choice but to start his own company.

A small local company offered Tobi a job.

Interested.

He must sign papers to join the local company.

He needed a work permit when he signed the documents.

Tobi had no work permit.

He was allowed to stay in Canada while applying for permanent residency.

“I wasn’t illegal in the country, but my state didn’t give me a work permit. I talked to a lawyer and he told me it’s going to take a while until I get a permanent residency.”

Tobi's lawyer told him he cannot get a work visa without permanent residence.

His lawyer said something else intriguing.

Tobis lawyer advised him to start a business.

Tobi declined this local company's job offer because of this.

Tobi considered opening an internet store with his technical skills.

He sold snowboards online.

“I was thinking of setting up an online store software because I figured that would exist and use it as a way to sell snowboards…make money while snowboarding and hopefully have a good life.”

What brought Tobi and his co-founder together, and how did he support Tobi?

Tobi lived with his girlfriend's parents.

In Ottawa, Tobi encounters Scott Lake.

Scott was Tobis girlfriend's family friend and worked for Tobi's future employer.

Scott and Tobi snowboarded.

Tobi pitched Scott his snowboard sales software idea.

Scott liked the idea.

They planned a business together.

“I was looking after the technology and Scott was dealing with the business side…It was Scott who ended up developing relationships with vendors and doing all the business set-up.”

Issues they ran into when attempting to launch their business online

Neither could afford a long-term lease.

That prompted their online business idea.

They would open a store.

Tobi anticipated opening an internet store in a week.

Tobi seeks open-source software.

Most existing software was pricey.

Tobi and Scott couldn't afford pricey software.

“In 2004, I was sitting in front of my computer absolutely stunned realising that we hadn’t figured out how to create software for online stores.”

They required software to:

• to upload snowboard images to the website.

• people to look up the types of snowboards that were offered on the website. There must be a search feature in the software.

• Online users transmit payments, and the merchant must receive them.

• notifying vendors of the recently received order.

No online selling software existed at the time.

Online credit card payments were difficult.

How did they advance the software while keeping expenses down?

Tobi and Scott needed money to start selling snowboards.

Tobi and Scott funded their firm with savings.

“We both put money into the company…I think the capital we had was around CAD 20,000(Canadian Dollars).”

Despite investing their savings.

They minimized costs.

They tried to conserve.

No office rental.

They worked in several coffee shops.

Tobi lived rent-free at his girlfriend's parents.

He installed software in coffee cafes.

How were the software issues handled?

Tobi found no online snowboard sales software.

Two choices remained:

1. Change your mind and try something else.

2. Use his programming expertise to produce something that will aid in the expansion of this company.

Tobi knew he was the sole programmer working on such a project from the start.

“I had this realisation that I’m going to be the only programmer who has ever worked on this, so I don’t have to choose something that lots of people know. I can choose just the best tool for the job…There is been this programming language called Ruby which I just absolutely loved ”

Ruby was open-source and only had Japanese documentation.

Latin is the source code.

Tobi used Ruby twice.

He assumed he could pick the tool this time.

Why not build with Ruby?

How did they find their first time operating a business?

Tobi writes applications in Ruby.

He wrote the initial software version in 2.5 months.

Tobi and Scott founded Snowdevil to sell snowboards.

Tobi coded for 16 hours a day.

His lifestyle was unhealthy.

He enjoyed pizza and coke.

“I would never recommend this to anyone, but at the time there was nothing more interesting to me in the world.”

Their initial purchase and encounter with it

Tobi worked in cafes then.

“I was working in a coffee shop at this time and I remember everything about that day…At some time, while I was writing the software, I had to type the email that the software would send to tell me about the order.”

Tobi recalls everything.

He checked the order on his laptop at the coffee shop.

Pennsylvanian ordered snowboard.

Tobi walked home and called Scott. Tobi told Scott their first order.

They loved the order.

How were people made aware about Snowdevil?

2004 was very different.

Tobi and Scott attempted simple website advertising.

Google AdWords was new.

Ad clicks cost 20 cents.

Online snowboard stores were scarce at the time.

Google ads propelled the snowdevil brand.

Snowdevil prospered.

They swiftly recouped their original investment in the snowboard business because to its high profit margin.

Tobi and Scott struggled with inventories.

“Snowboards had really good profit margins…Our biggest problem was keeping inventory and getting it back…We were out of stock all the time.”

Selling snowboards returned their investment and saved them money.

They did not appoint a business manager.

They accomplished everything alone.

Sales dipped in the spring, but something magical happened.

Spring sales plummeted.

They considered stocking different boards.

They naturally wanted to add boards and grow the business.

However, magic occurred.

Tobi coded and improved software while running Snowdevil.

He modified software constantly. He wanted speedier software.

He experimented to make the software more resilient.

Tobi received emails requesting the Snowdevil license.

They intended to create something similar.

“I didn’t stop programming, I was just like Ok now let me try things, let me make it faster and try different approaches…Increasingly I got people sending me emails and asking me If I would like to licence snowdevil to them. People wanted to start something similar.”

Software or skateboards, your choice

Scott and Tobi had to choose a hobby in 2005.

They might sell alternative boards or use software.

The software was a no-brainer from demand.

Daniel Weinand is invited to join Tobi's business.

Tobis German best friend is Daniel.

Tobi and Scott chose to use the software.

Tobi and Scott kept the software service.

Tobi called Daniel to invite him to Canada to collaborate.

Scott and Tobi had quit snowboarding until then.

How was Shopify launched, and whence did the name come from?

The three chose Shopify.

Named from two words.

First:

• Shop

Final part:

• Simplify

Shopify

Shopify's crew has always had one goal:

• creating software that would make it simple and easy for people to launch online storefronts.

Launched Shopify after raising money for the first time.

Shopify began fundraising in 2005.

First, they borrowed from family and friends.

They needed roughly $200k to run the company efficiently.

$200k was a lot then.

When questioned why they require so much money. Tobi told them to trust him with their goals. The team raised seed money from family and friends.

Shopify.com has a landing page. A demo of their goal was on the landing page.

In 2006, Shopify had about 4,000 emails.

Shopify rented an Ottawa office.

“We sent a blast of emails…Some people signed up just to try it out, which was exciting.”

How things developed after Scott left the company

Shopify co-founder Scott Lake left in 2008.

Scott was CEO.

“He(Scott) realized at some point that where the software industry was going, most of the people who were the CEOs were actually the highly technical person on the founding team.”

Scott leaving the company worried Tobi.

Tobis worried about finding a new CEO.

To Tobi:

A great VC will have the network to identify the perfect CEO for your firm.

Tobi started visiting Silicon Valley to meet with venture capitalists to recruit a CEO.

Initially visiting Silicon Valley

Tobi came to Silicon Valley to start a 20-person company.

This company creates eCommerce store software.

Tobi never wanted a big corporation. He desired a fulfilling existence.

“I stayed in a hostel in the Bay Area. I had one roommate who was also a computer programmer. I bought a bicycle on Craiglist. I was there for a week, but ended up staying two and a half weeks.”

Tobi arrived unprepared.

When venture capitalists asked him business questions.

He answered few queries.

Tobi didn't comprehend VC meetings' terminology.

He wrote the terms down and looked them up.

Some were fascinated after he couldn't answer all these queries.

“I ended up getting the kind of term sheets people dream about…All the offers were conditional on moving our company to Silicon Valley.”

Canada received Tobi.

He wanted to consult his team before deciding. Shopify had five employees at the time.

2008.

A global recession greeted Tobi in Canada. The recession hurt the market.

His term sheets were useless.

The economic downturn in the world provided Shopify with a fantastic opportunity.

The global recession caused significant job losses.

Fired employees had several ideas.

They wanted online stores.

Entrepreneurship was desired. They wanted to quit work.

People took risks and tried new things during the global slump.

Shopify subscribers skyrocketed during the recession.

“In 2009, the company reached neutral cash flow for the first time…We were in a position to think about long-term investments, such as infrastructure projects.”

Then, Tobi Lutke became CEO.

How did Tobi perform as the company's CEO?

“I wasn’t good. My team was very patient with me, but I had a lot to learn…It’s a very subtle job.”

2009–2010.

Tobi limited the company's potential.

He deliberately restrained company growth.

Tobi had one costly problem:

• Whether Shopify is a venture or a lifestyle business.

The company's annual revenue approached $1 million.

Tobi battled with the firm and himself despite good revenue.

His wife was supportive, but the responsibility was crushing him.

“It’s a crushing responsibility…People had families and kids…I just couldn’t believe what was going on…My father-in-law gave me money to cover the payroll and it was his life-saving.”

Throughout this trip, everyone supported Tobi.

They believed it.

$7 million in donations received

Tobi couldn't decide if this was a lifestyle or a business.

Shopify struggled with marketing then.

Later, Tobi tried 5 marketing methods.

He told himself that if any marketing method greatly increased their growth, he would call it a venture, otherwise a lifestyle.

The Shopify crew brainstormed and voted on marketing concepts.

Tested.

“Every single idea worked…We did Adwords, published a book on the concept, sponsored a podcast and all the ones we tracked worked.”

To Silicon Valley once more

Shopify marketing concepts worked once.

Tobi returned to Silicon Valley to pitch investors.

He raised $7 million, valuing Shopify at $25 million.

All investors had board seats.

“I find it very helpful…I always had a fantastic relationship with everyone who’s invested in my company…I told them straight that I am not going to pretend I know things, I want you to help me.”

Tobi developed skills via running Shopify.

Shopify had 20 employees.

Leaving his wife's parents' home

Tobi left his wife's parents in 2014.

Tobi had a child.

Shopify has 80,000 customers and 300 staff in 2013.

Public offering in 2015

Shopify investors went public in 2015.

Shopify powers 4.1 million e-Commerce sites.

Shopify stores are 65% US-based.

It is currently valued at $48 billion.

A fascinating business conversation.

This week at Web Summit, I had mentor hour.

Mentor hour connects startups with experienced entrepreneurs.

The YC-selected founder who mentored me had grown his company to $100 million in 18 months.

I had 45 minutes to question him.

I've compiled this.

Context

Founder's name is Zack.

After working in private equity, Zack opted to acquire an MBA.

Surrounded by entrepreneurs at a prominent school, he decided to become one himself.

Unsure how to proceed, he bet on two horses.

On one side, he received an offer from folks who needed help running their startup owing to lack of time. On the other hand, he had an idea for a SaaS to start himself.

He just needed to validate it.

Validating

Since Zack's proposal helped companies, he contacted university entrepreneurs for comments.

He contacted university founders.

Once he knew he'd correctly identified the problem and that people were willing to pay to address it, he started developing.

He earned $100k in a university entrepreneurship competition.

His plan was evident by then.

The other startup's founders saw his potential and granted him $400k to launch his own SaaS.

Hiring

He started looking for a tech co-founder because he lacked IT skills.

He interviewed dozens and picked the finest.

As he didn't want to wait for his program to be ready, he contacted hundreds of potential clients and got 15 letters of intent promising they'd join up when it was available.

YC accepted him by then.

He had enough positive signals to raise.

Raising

He didn't say how many VCs he called, but he indicated 50 were interested.

He jammed meetings into two weeks to generate pressure and encourage them to invest.

Seed raise: $11 million.

Selling

His objective was to contact as many entrepreneurs as possible to promote his product.

He first contacted startups by scraping CrunchBase data.

Once he had more money, he started targeting companies with ZoomInfo.

His VC urged him not to hire salespeople until he closed 50 clients himself.

He closed 100 and hired a CRO through a headhunter.

Scaling

Three persons started the business.

1. He primarily works in sales.

2. Coding the product was done by his co-founder.

3. Another person performing operational duties.

He regretted recruiting the third co-founder, who was ineffective (could have hired an employee instead).

He wanted his company to be big, so he hired two young marketing people from a competing company.

After validating several marketing channels, he chose PR.

$100 Million and under

He developed a sales team and now employs 30 individuals.

He raised a $100 million Series A.

Additionally, he stated

• He’s been rejected a lot. Like, a lot.

• Two great books to read: Steve Jobs by Isaacson, and Why Startups Fail by Tom Eisenmann.

• The best skill to learn for non-tech founders is “telling stories”, which means sales. A founder’s main job is to convince: co-founders, employees, investors, and customers. Learn code, or learn sales.

Conclusion

I often read about these stories but hardly take them seriously.

Zack was amazing.

Three things about him stand out:

1. His vision. He possessed a certain amount of fire.

2. His vitality. The man had a lot of enthusiasm and spoke quickly and decisively. He takes no chances and pushes the envelope in all he does.

3. His Rolex.

He didn't do all this in 18 months.

Not really.

He couldn't launch his company without private equity experience.

These accounts disregard entrepreneurs' original knowledge.

Hormozi will tell you how he founded Gym Launch, but he won't tell you how he had a gym first, how he worked at uni to pay for his gym, or how he went to the gym and learnt about fitness, which gave him the idea to open his own.

Nobody knows nothing. If you scale quickly, it's probable because you gained information early.

Lincoln said, "Give me six hours to chop down a tree, and I'll spend four sharpening the axe."

Sharper axes cut trees faster.

He’s not the President. Stop treating him like one.

Elon Musk tweeted from Qatar, where he was watching the World Cup Final with Jared Kushner.

Musk's subsequent Tweets were as normal, basic, and bland as anyone's from a World Cup Final: It's depressing to see the world's richest man looking at his phone during a grand ceremony. Rich guy goes to rich guy event didn't seem important.

Before Musk posted his should-I-step-down-at-Twitter poll, CNN ran a long segment asking if it was hypocritical for him to reveal his real-time location after defending his (very dumb) suspension of several journalists for (supposedly) revealing his assassination coordinates by linking to a site that tracks Musks private jet. It was hard to ignore CNN's hypocrisy: It covered Musk as Twitter CEO like President Trump. EVERY TRUMP STORY WAS BASED ON HIM SAYING X, THEN DOING Y. Trump would do something horrific, lie about it, then pretend it was fine, then condemn a political rival who did the same thing, be called hypocritical, and so on. It lasted four years. Exhausting.

It made sense because Trump was the President of the United States. The press's main purpose is to relentlessly cover and question the president.

It's strange to say this out. Twitter isn't America. Elon Musk isn't a president. He maintains a money-losing social media service to harass and mock people he doesn't like. Treating Musk like Trump, as if he should be held accountable like Trump, shows a startling lack of perspective. Some journalists treat Twitter like a country.

The compulsive, desperate way many journalists utilize the site suggests as much. Twitter isn't the town square, despite popular belief. It's a place for obsessives to meet and converse. Journalists say they're breaking news. Their careers depend on it. They can argue it's a public service. Nope. It's a place lonely people go to speak all day. Twitter. So do journalists, Trump, and Musk. Acting as if it has a greater purpose, as if it's impossible to break news without it, or as if the republic is in peril is ludicrous. Only 23% of Americans are on Twitter, while 25% account for 97% of Tweets. I'd think a large portion of that 25% are journalists (or attention addicts) chatting to other journalists. Their loudness makes Twitter seem more important than it is. Nope. It's another stupid website. They were there before Twitter; they will be there after Twitter. It’s just a website. We can all get off it if we want. Most of us aren’t even on it in the first place.

Musk is a website-owner. No world leader. He's not as accountable as Trump was. Musk is cable news's primary character now that Trump isn't (at least for now). Becoming a TV news anchor isn't as significant as being president. Elon Musk isn't as important as we all pretend, and Twitter isn't even close. Twitter is a dumb website, Elon Musk is a rich guy going through a midlife crisis, and cable news is lazy because its leaders thought the entire world was on Twitter and are now freaking out that their playground is being disturbed.

I’ve said before that you need to leave Twitter, now. But even if you’re still on it, we need to stop pretending it matters more than it does. It’s a site for lonely attention addicts, from the man who runs it to the journalists who can’t let go of it. It’s not a town square. It’s not a country. It’s not even a successful website. Let’s stop pretending any of it’s real. It’s not.

Deep learning-based generative art is being researched. As usual, self-learning is better. Some models, like OpenAI's DALL-E 2, require registration and can only be used online, but others can be used locally, which is usually more enjoyable for curious users. I'll demonstrate the Stable Diffusion model's operation on a standard PC.

Let’s get started.

What It Does

Stable Diffusion uses numerous components:

• A generative model trained to produce images is called a diffusion model. The model is incrementally improving the starting data, which is only random noise. The model has an image, and while it is being trained, the reversed process is being used to add noise to the image. Being able to reverse this procedure and create images from noise is where the true magic is (more details and samples can be found in the paper).

• An internal compressed representation of a latent diffusion model, which may be altered to produce the desired images, is used (more details can be found in the paper). The capacity to fine-tune the generation process is essential because producing pictures at random is not very attractive (as we can see, for instance, in Generative Adversarial Networks).

• A neural network model called CLIP (Contrastive Language-Image Pre-training) is used to translate natural language prompts into vector representations. This model, which was trained on 400,000,000 image-text pairs, enables the transformation of a text prompt into a latent space for the diffusion model in the scenario of stable diffusion (more details in that paper).

This figure shows all data flow:

The weights file size for Stable Diffusion model v1 is 4 GB and v2 is 5 GB, making the model quite huge. The v1 model was trained on 256x256 and 512x512 LAION-5B pictures on a 4,000 GPU cluster using over 150.000 NVIDIA A100 GPU hours. The open-source pre-trained model is helpful for us. And we will.

Install

Before utilizing the Python sources for Stable Diffusion v1 on GitHub, we must install Miniconda (assuming Git and Python are already installed):

wget https://repo.anaconda.com/miniconda/Miniconda3-py39_4.12.0-Linux-x86_64.sh
chmod +x Miniconda3-py39_4.12.0-Linux-x86_64.sh
./Miniconda3-py39_4.12.0-Linux-x86_64.sh
conda update -n base -c defaults conda

Install the source and prepare the environment:

git clone https://github.com/CompVis/stable-diffusion
cd stable-diffusion
conda env create -f environment.yaml
conda activate ldm
pip3 install transformers --upgrade

Download the pre-trained model weights next. HiggingFace has the newest checkpoint sd-v14.ckpt (a download is free but registration is required). Put the file in the project folder and have fun:

python3 scripts/txt2img.py --prompt "hello world" --plms --ckpt sd-v1-4.ckpt --skip_grid --n_samples 1

Almost. The installation is complete for happy users of current GPUs with 12 GB or more VRAM. RuntimeError: CUDA out of memory will occur otherwise. Two solutions exist.

Running the optimized version

Try optimizing first. After cloning the repository and enabling the environment (as previously), we can run the command:

python3 optimizedSD/optimized_txt2img.py --prompt "hello world" --ckpt sd-v1-4.ckpt --skip_grid --n_samples 1

Stable Diffusion worked on my visual card with 8 GB RAM (alas, I did not behave well enough to get NVIDIA A100 for Christmas, so 8 GB GPU is the maximum I have;).

Running Stable Diffusion without GPU

If the GPU does not have enough RAM or is not CUDA-compatible, running the code on a CPU will be 20x slower but better than nothing. This unauthorized CPU-only branch from GitHub is easiest to obtain. We may easily edit the source code to use the latest version. It's strange that a pull request for that was made six months ago and still hasn't been approved, as the changes are simple. Readers can finish in 5 minutes:

• Replace if attr.device!= torch.device(cuda) with if attr.device!= torch.device(cuda) and torch.cuda.is available at line 20 of ldm/models/diffusion/ddim.py ().

• Replace if attr.device!= torch.device(cuda) with if attr.device!= torch.device(cuda) and torch.cuda.is available in line 20 of ldm/models/diffusion/plms.py ().

• Replace device=cuda in lines 38, 55, 83, and 142 of ldm/modules/encoders/modules.py with device=cuda if torch.cuda.is available(), otherwise cpu.

• Replace model.cuda() in scripts/txt2img.py line 28 and scripts/img2img.py line 43 with if torch.cuda.is available(): model.cuda ().

Run the script again.

Testing

Test the model. Text-to-image is the first choice. Test the command line example again:

python3 scripts/txt2img.py --prompt "hello world" --plms --ckpt sd-v1-4.ckpt --skip_grid --n_samples 1

The slow generation takes 10 seconds on a GPU and 10 minutes on a CPU. Final image:

Hello world is dull and abstract. Try a brush-wielding hamster. Why? Because we can, and it's not as insane as Napoleon's cat. Another image:

Generating an image from a text prompt and another image is interesting. I made this picture in two minutes using the image editor (sorry, drawing wasn't my strong suit):

I can create an image from this drawing:

python3 scripts/img2img.py --prompt "A bird is sitting on a tree branch" --ckpt sd-v1-4.ckpt --init-img bird.png --strength 0.8

It was far better than my initial drawing:

I hope readers understand and experiment.

Stable Diffusion UI

Developers love the command line, but regular users may struggle. Stable Diffusion UI projects simplify image generation and installation. Simple usage:

• Unpack the ZIP after downloading it from https://github.com/cmdr2/stable-diffusion-ui/releases. Linux and Windows are compatible with Stable Diffusion UI (sorry for Mac users, but those machines are not well-suitable for heavy machine learning tasks anyway;).

• Start the script.

Done. The web browser UI makes configuring various Stable Diffusion features (upscaling, filtering, etc.) easy:

V2.1 of Stable Diffusion

I noticed the notification about releasing version 2.1 while writing this essay, and it was intriguing to test it. First, compare version 2 to version 1:

• alternative text encoding. The Contrastive LanguageImage Pre-training (CLIP) deep learning model, which was trained on a significant number of text-image pairs, is used in Stable Diffusion 1. The open-source CLIP implementation used in Stable Diffusion 2 is called OpenCLIP. It is difficult to determine whether there have been any technical advancements or if legal concerns were the main focus. However, because the training datasets for the two text encoders were different, the output results from V1 and V2 will differ for the identical text prompts.

• a new depth model that may be used to the output of image-to-image generation.

• a revolutionary upscaling technique that can quadruple the resolution of an image.

• Generally higher resolution Stable Diffusion 2 has the ability to produce both 512x512 and 768x768 pictures.

The Hugging Face website offers a free online demo of Stable Diffusion 2.1 for code testing. The process is the same as for version 1.4. Download a fresh version and activate the environment:

conda deactivate  
conda env remove -n ldm  # Use this if version 1 was previously installed
git clone https://github.com/Stability-AI/stablediffusion
cd stablediffusion
conda env create -f environment.yaml
conda activate ldm

Hugging Face offers a new weights ckpt file.

The Out of memory error prevented me from running this version on my 8 GB GPU. Version 2.1 fails on CPUs with the slow conv2d cpu not implemented for Half error (according to this GitHub issue, the CPU support for this algorithm and data type will not be added). The model can be modified from half to full precision (float16 instead of float32), however it doesn't make sense since v1 runs up to 10 minutes on the CPU and v2.1 should be much slower. The online demo results are visible. The same hamster painting with a brush prompt yielded this result:

It looks different from v1, but it functions and has a higher resolution.

The superresolution.py script can run the 4x Stable Diffusion upscaler locally (the x4-upscaler-ema.ckpt weights file should be in the same folder):

python3 scripts/gradio/superresolution.py configs/stable-diffusion/x4-upscaling.yaml x4-upscaler-ema.ckpt

This code allows the web browser UI to select the image to upscale:

The copy-paste strategy may explain why the upscaler needs a text prompt (and the Hugging Face code snippet does not have any text input as well). I got a GPU out of memory error again, although CUDA can be disabled like v1. However, processing an image for more than two hours is unlikely:

Stable Diffusion Limitations

When we use the model, it's fun to see what it can and can't do. Generative models produce abstract visuals but not photorealistic ones. This fundamentally limits The generative neural network was trained on text and image pairs, but humans have a lot of background knowledge about the world. The neural network model knows nothing. If someone asks me to draw a Chinese text, I can draw something that looks like Chinese but is actually gibberish because I never learnt it. Generative AI does too! Humans can learn new languages, but the Stable Diffusion AI model includes only language and image decoder brain components. For instance, the Stable Diffusion model will pull NO WAR banner-bearers like this:

V1:

V2.1:

The shot shows text, although the model never learned to read or write. The model's string tokenizer automatically converts letters to lowercase before generating the image, so typing NO WAR banner or no war banner is the same.

I can also ask the model to draw a gorgeous woman:

V1:

V2.1:

The first image is gorgeous but physically incorrect. A second one is better, although it has an Uncanny valley feel. BTW, v2 has a lifehack to add a negative prompt and define what we don't want on the image. Readers might try adding horrible anatomy to the gorgeous woman request.

If we ask for a cartoon attractive woman, the results are nice, but accuracy doesn't matter:

V1:

V2.1:

Another example: I ordered a model to sketch a mouse, which looks beautiful but has too many legs, ears, and fingers:

V1:

V2.1: improved but not perfect.

V1 produces a fun cartoon flying mouse if I want something more abstract:

I tried multiple times with V2.1 but only received this:

The image is OK, but the first version is closer to the request.

Stable Diffusion struggles to draw letters, fingers, etc. However, abstract images yield interesting outcomes. A rural landscape with a modern metropolis in the background turned out well:

V1:

V2.1:

Generative models help make paintings too (at least, abstract ones). I searched Google Image Search for modern art painting to see works by real artists, and this was the first image:

I typed "abstract oil painting of people dancing" and got this:

V1:

V2.1:

It's a different style, but I don't think the AI-generated graphics are worse than the human-drawn ones.

The AI model cannot think like humans. It thinks nothing. A stable diffusion model is a billion-parameter matrix trained on millions of text-image pairs. I input "robot is creating a picture with a pen" to create an image for this post. Humans understand requests immediately. I tried Stable Diffusion multiple times and got this:

This great artwork has a pen, robot, and sketch, however it was not asked. Maybe it was because the tokenizer deleted is and a words from a statement, but I tried other requests such robot painting picture with pen without success. It's harder to prompt a model than a person.

I hope Stable Diffusion's general effects are evident. Despite its limitations, it can produce beautiful photographs in some settings. Readers who want to use Stable Diffusion results should be warned. Source code examination demonstrates that Stable Diffusion images feature a concealed watermark (text StableDiffusionV1 and SDV2) encoded using the invisible-watermark Python package. It's not a secret, because the official Stable Diffusion repository's test watermark.py file contains a decoding snippet. The put watermark line in the txt2img.py source code can be removed if desired. I didn't discover this watermark on photographs made by the online Hugging Face demo. Maybe I did something incorrectly (but maybe they are just not using the txt2img script on their backend at all).

Conclusion

The Stable Diffusion model was fascinating. As I mentioned before, trying something yourself is always better than taking someone else's word, so I encourage readers to do the same (including this article as well;).

Is Generative AI a game-changer? My humble experience tells me:

• I think that place has a lot of potential. For designers and artists, generative AI can be a truly useful and innovative tool. Unfortunately, it can also pose a threat to some of them since if users can enter a text field to obtain a picture or a website logo in a matter of clicks, why would they pay more to a different party? Is it possible right now? unquestionably not yet. Images still have a very poor quality and are erroneous in minute details. And after viewing the image of the stunning woman above, models and fashion photographers may also unwind because it is highly unlikely that AI will replace them in the upcoming years.

• Today, generative AI is still in its infancy. Even 768x768 images are considered to be of a high resolution when using neural networks, which are computationally highly expensive. There isn't an AI model that can generate high-resolution photographs natively without upscaling or other methods, at least not as of the time this article was written, but it will happen eventually.

• It is still a challenge to accurately represent knowledge in neural networks (information like how many legs a cat has or the year Napoleon was born). Consequently, AI models struggle to create photorealistic photos, at least where little details are important (on the other side, when I searched Google for modern art paintings, the results are often even worse;).

• When compared to the carefully chosen images from official web pages or YouTube reviews, the average output quality of a Stable Diffusion generation process is actually less attractive because to its high degree of randomness. When using the same technique on their own, consumers will theoretically only view those images as 1% of the results.

Anyway, it's exciting to witness this area's advancement, especially because the project is open source. Google's Imagen and DALL-E 2 can also produce remarkable findings. It will be interesting to see how they progress.