How a misplaced item may change your outlook

Losing your wallet means life stops. Money vanishes. No credit. Your identity is unverifiable. As you check your pockets for the missing object, you can't drive. You can't borrow a library book.

Last seen? intuitively. Every kid asks this, including yours. However, you know where you lost it: On the Providence River cycling trail. While pedaling vigorously, the wallet dropped out of your back pocket and onto the pavement.

A woman you know—your son's art teacher—says it will be returned. Faith.

You want that faith. Losing a wallet is all-consuming. You must presume it has been stolen and is being used to buy every diamond and non-fungible token on the market. Your identity may have been used to open bank accounts and fake passports. Because he used your license address, a ski mask-wearing man may be driving slowly past your house.

As you delete yourself by canceling cards, these images run through your head. You wait in limbo for replacements. Digital text on the DMV website promises your new license will come within 60 days and be approved by local and state law enforcement. In the following two months, your only defense is a screenshot.

Your wallet was ordinary. A worn, overstuffed leather rectangle. You understand how tenuous your existence has always been since you've never lost a wallet. You barely breathe without your documents.

Ironically, you wore a wallet-belt chain. You adored being a 1993 slacker for 15 years. Your wife just convinced you last year that your office job wasn't professional. You nodded and hid the chain.

Never lost your wallet. Until now.

Angry. Feeling stupid. How could you drop something vital? Why? Is the world cruel? No more dumb luck. You're always one pedal-stroke from death.

Then you get a call: We have your wallet.

Local post office, not cops.

The clerk said someone returned it. Due to trying to identify you, it's a chaos. It has your cards but no cash.

Your automobile screeches down the highway. You yell at the windshield, amazed. Submitted. Art teacher was right. Have some trust.

You thank the postmaster. You ramble through the story. The clerk doesn't know the customer, simply a neighborhood Good Samaritan. You wish you could thank that person for lifting your spirits.

You get home, beaming with gratitude. You thumb through your wallet, amazed that it’s all intact. Then you dig out your chain and reattach it.

Because even faith could use a little help.

Are you one of these Zoom dogs?

The Person Who Is Apparently Always on Mute

Waffles thinks he can overpower the mute button by shouting loudly.

The person who believed their camera to be off

Barkley's used to remote work, but he hasn't mastered the "Stop Video" button. Everyone is affected.

Who is driving for some reason, exactly?

Why is Pumpkin always late? Who knows? Shouldn't she be driving? If you could hear her over the freeway, she'd answer these questions.

The Person With the Amazing Bookcase

Cicero likes to use SAT-words like "leverage" and "robust" in Zoom sessions, presumably from all the books he wants you to see behind him.

The Individual Who Is Unnecessarily Dressed

We hope Bandit is going somewhere beautiful after this meeting, or else he neglected the quarterly earnings report and is overcompensating to distract us.

The person who works through lunch in between zoom calls

Barksworth has back-to-back meetings all day, so you can watch her eat while she talks.

The Person Who Is A Little Too Comfy

Hercules thinks Zoom meetings happen between sleeps. He'd appreciate everyone speaking more quietly.

The Person Who Answered the Phone Outside

Frisbee has a gorgeous backyard and lives in a place with great weather year-round, and she wants you to think about that during the daily team huddle.

Who Wants You to Pay Attention to Their Pet

Snickers hasn't listened to you in 20 minutes unless you tell her how cute her kitten is.

One who is, for some reason, positioned incorrectly on the screen

Nelson's meetings consist primarily of attempting to figure out how he positioned his laptop so absurdly.

The person who says too many goodbyes

Zeus waves farewell like it's your first day of school while everyone else searches for the "Leave Meeting" button. It's nice.

He who has a poor internet connection

Ziggy's connectivity problems continue... She gives a long speech as everyone waits awkwardly to inform her they missed it.

The Clearly Multitasking Person

Tinkerbell can play fetch during the monthly staff meeting if she works from home, but that's not a good idea.

The Person Using Zoom as a Makeup and Hair Mirror

If Gail and Bob knew Zoom had a "hide self view" option, they'd be distraught.

The person who feels at ease with simply leaving

Rusty bails when a Zoom conference is over. Rusty's concept is decent.

Google announced 6% layoffs recently (or 12,000 people). This aligns it with most tech companies. A publicly contrite CEO explained that they had overhired during the COVID-19 pandemic boom and had to address it, but they were sorry and took full responsibility. I thought this was "bullshit" too. Meta, Amazon, Microsoft, and others must feel similarly. I spent 10 years at Google, and these things don't reflect well on the company's leaders.

All publicly listed companies have a fiduciary duty to act in the best interests of their shareholders. Dodge vs. Ford Motor Company established this (1919). Henry Ford wanted to reduce shareholder payments to offer cheaper cars and better wages. Ford stated.

My ambition is to employ still more men, to spread the benefits of this industrial system to the greatest possible number, to help them build up their lives and their homes. To do this we are putting the greatest share of our profits back in the business.

The Dodge brothers, who owned 10% of Ford, opposed this and sued Ford for the payments to start their own company. They won, preventing Ford from raising prices or salaries. If you have a vocal group of shareholders with the resources to sue you, you must prove you are acting in their best interests. Companies prioritize shareholders. Giving activist investors a stick to threaten you almost enshrines short-term profit over long-term thinking.

This underpins Google's current issues. Institutional investors who can sue Google see it as a wasteful company they can exploit. That doesn't mean you have to maximize profits (thanks to those who pointed out my ignorance of US corporate law in the comments and on HN), but it allows pressure. I feel for those navigating this. This is about unrestrained capitalism.

When Google went public, Larry Page and Sergey Brin knew the risks and worked hard to keep control. In their Founders' Letter to investors, they tried to set expectations for the company's operations.

Our long-term focus as a private company has paid off. Public companies do the same. We believe outside pressures lead companies to sacrifice long-term opportunities to meet quarterly market expectations.

The company has transformed since that letter. The company has nearly 200,000 full-time employees and a trillion-dollar market cap. Large investors have bought company stock because it has been a good long-term bet. Why are they restless now?

Other big tech companies emerged and fought for top talent. This has caused rising compensation packages. Google has also grown rapidly (roughly 22,000 people hired to the end of 2022). At $300,000 median compensation, those 22,000 people added $6.6 billion in salary overheads in 2022. Exorbitant. If the company still makes $16 billion every quarter, maybe not. Investors wonder if this value has returned.

Investors are right. Google uses people wastefully. However, by bluntly reducing headcount, they're not addressing the root causes and hurting themselves. No studies show that downsizing this way boosts productivity. There is plenty of evidence that they'll lose out because people will be risk-averse and distrust their leadership.

The company's approach also stinks. Finding out that you no longer have a job because you can’t log in anymore (sometimes in cases where someone is on call for protecting your production systems) is no way to fire anyone. Being with a narcissistic sociopath is like being abused. First, you receive praise and fancy perks for making the cut. You're fired by text and ghosted. You're told to appreciate the generous severance package. This firing will devastate managers and teams. This type of firing will take years to recover self-esteem. Senior management contributed to this. They chose the expedient answer, possibly by convincing themselves they were managing risk and taking the Macbeth approach of “If it were done when ’tis done, then ’twere well It were done quickly”.

Recap. Google's leadership did a stupid thing—mass firing—in a stupid way. How do we get rid of enough people to make investors happier? and "have 6% less people." Empathetic leaders should not emulate Elon Musk. There is no humane way to fire 12,000 people, but there are better ways. Why is Google so wasteful?

Ambition answers this. There aren't enough VP positions for a group of highly motivated, ambitious, and (increasingly) ruthless people. I’ve loitered around the edges of this world and a large part of my value was to insulate my teams from ever having to experience it. It’s like Game of Thrones played out through email and calendar and over video call.

Your company must look a certain way to be promoted to director or higher. You need the right people at the right levels under you. Long-term, growing your people will naturally happen if you're working on important things. This takes time, and you're never more than 6–18 months from a reorg that could start you over. Ambitious people also tend to be impatient. So, what do you do?

Hiring and vanity projects. To shape your company, you hire at the right levels. You value vanity metrics like active users over product utility. Your promo candidates get through by subverting the promotion process. In your quest for growth, you avoid performance managing people out. You avoid confronting toxic peers because you need their support for promotion. Your cargo cult gets you there.

Its ease makes Google wasteful. Since they don't face market forces, the employees don't see it as a business. Why would you do when the ads business is so profitable? Complacency causes senior leaders to prioritize their own interests. Empires collapse. Personal ambition often trumped doing the right thing for users, the business, or employees. Leadership's ambition over business is the root cause. Vanity metrics, mass hiring, and vague promises have promoted people to VP. Google goes above and beyond to protect senior leaders.

The decision-makers and beneficiaries are not the layoffees. Stock price increase beneficiaries. The people who will post on LinkedIn how it is about misjudging the market and how they’re so sorry and take full responsibility. While accumulating wealth, the dark room dwellers decide who stays and who goes. The billionaire investors. Google should start by addressing its bloated senior management, but — as they say — turkeys don't vote for Christmas. It should examine its wastefulness and make tough choices to fix it. A 6% cut is a blunt tool that admits you're not running your business properly. why aren’t the people running the business the ones shortly to be entering the job market?

This won't fix Google's wastefulness. The executives may never regain trust after their approach. Suppressed creativity. Business won't improve. Google will have lost its founding vision and us all. Large investors know they can force Google's CEO to yield. The rich will get richer and rationalize leaving 12,000 people behind. Cycles repeat.

It doesn’t have to be this way. In 2013, Nintendo's CEO said he wouldn't fire anyone for shareholders. Switch debuted in 2017. Nintendo's stock has increased by nearly five times, or 19% a year (including the drop most of the stock market experienced last year). Google wasted 12,000 talented people. To please rich people.

Sam Bankman Fried (SDBF) is a legendary con man. But the NYT might not tell you that...

Since SBF's empire was revealed to be a lie, mainstream news organizations and commentators have failed to give readers a straightforward assessment. The New York Times and Wall Street Journal have uncovered many key facts about the scandal, but they have also soft-peddled Bankman-Fried's intent and culpability.

It's clear that the FTX crypto exchange and Alameda Research committed fraud to steal money from users and investors. That’s why a recent New York Times interview was widely derided for seeming to frame FTX’s collapse as the result of mismanagement rather than malfeasance. A Wall Street Journal article lamented FTX's loss of charitable donations, bolstering Bankman's philanthropic pose. Matthew Yglesias, court chronicler of the neoliberal status quo, seemed to whitewash his own entanglements by crediting SBF's money with helping Democrats in 2020 – sidestepping the likelihood that the money was embezzled.

Many outlets have called what happened to FTX a "bank run" or a "run on deposits," but Bankman-Fried insists the company was overleveraged and disorganized. Both attempts to frame the fallout obscure the core issue: customer funds misused.

Because banks lend customer funds to generate returns, they can experience "bank runs." If everyone withdraws at once, they can experience a short-term cash crunch but there won't be a long-term problem.

Crypto exchanges like FTX aren't banks. They don't do bank-style lending, so a withdrawal surge shouldn't strain liquidity. FTX promised customers it wouldn't lend or use their crypto.

Alameda's balance sheet blurs SBF's crypto empire.

The funds were sent to Alameda Research, where they were apparently gambled away. This is massive theft. According to a bankruptcy document, up to 1 million customers could be affected.

In less than a month, reporting and the bankruptcy process have uncovered a laundry list of decisions and practices that would constitute financial fraud if FTX had been a U.S.-regulated entity, even without crypto-specific rules. These ploys may be litigated in U.S. courts if they enabled the theft of American property.

The list is very, very long.

The many crimes of Sam Bankman-Fried and FTX

At the heart of SBF's fraud are the deep and (literally) intimate ties between FTX and Alameda Research, a hedge fund he co-founded. An exchange makes money from transaction fees on user assets, but Alameda trades and invests its own funds.

Bankman-Fried called FTX and Alameda "wholly separate" and resigned as Alameda's CEO in 2019. The two operations were closely linked. Bankman-Fried and Alameda CEO Caroline Ellison were romantically linked.

These circumstances enabled SBF's sin.  Within days of FTX's first signs of weakness, it was clear the exchange was funneling customer assets to Alameda for trading, lending, and investing. Reuters reported on Nov. 12 that FTX sent $10 billion to Alameda. As much as $2 billion was believed to have disappeared after being sent to Alameda. Now the losses look worse.

It's unclear why those funds were sent to Alameda or when Bankman-Fried betrayed his depositors. On-chain analysis shows most FTX to Alameda transfers occurred in late 2021, and bankruptcy filings show both lost $3.7 billion in 2021.

SBF's companies lost millions before the 2022 crypto bear market. They may have stolen funds before Terra and Three Arrows Capital, which killed many leveraged crypto players.

FTT loans and prints

CoinDesk's report on Alameda's FTT holdings ignited FTX and Alameda Research. FTX created this instrument, but only a small portion was traded publicly; FTX and Alameda held the rest. These holdings were illiquid, meaning they couldn't be sold at market price. Bankman-Fried valued its stock at the fictitious price.

FTT tokens were reportedly used as collateral for loans, including FTX loans to Alameda. Close ties between FTX and Alameda made the FTT token harder or more expensive to use as collateral, reducing the risk to customer funds.

This use of an internal asset as collateral for loans between clandestinely related entities is similar to Enron's 1990s accounting fraud. These executives served 12 years in prison.

Alameda's margin liquidation exemption

Alameda Research had a "secret exemption" from FTX's liquidation and margin trading rules, according to legal filings by FTX's new CEO.

FTX, like other crypto platforms and some equity or commodity services, offered "margin" or loans for trades. These loans are usually collateralized, meaning borrowers put up other funds or assets. If a margin trade loses enough money, the exchange will sell the user's collateral to pay off the initial loan.

Keeping asset markets solvent requires liquidating bad margin positions. Exempting Alameda would give it huge advantages while exposing other FTX users to hidden risks. Alameda could have kept losing positions open while closing out competitors. Alameda could lose more on FTX than it could pay back, leaving a hole in customer funds.

The exemption is criminal in multiple ways. FTX was fraudulently marketed overall. Instead of a level playing field, there were many customers.

Above them all, with shotgun poised, was Alameda Research.

Alameda front-running FTX listings

Argus says there's circumstantial evidence that Alameda Research had insider knowledge of FTX's token listing plans. Alameda was able to buy large amounts of tokens before the listing and sell them after the price bump.

If true, these claims would be the most brazenly illegal of Alameda and FTX's alleged shenanigans. Even if the tokens aren't formally classified as securities, insider trading laws may apply.

In a similar case this year, an OpenSea employee was charged with wire fraud for allegedly insider trading. This employee faces 20 years in prison for front-running monkey JPEGs.

Huge loans to executives

Alameda Research reportedly lent FTX executives $4.1 billion, including massive personal loans. Bankman-Fried received $1 billion in personal loans and $2.3 billion for an entity he controlled, Paper Bird. Nishad Singh, director of engineering, was given $543 million, and FTX Digital Markets co-CEO Ryan Salame received $55 million.

FTX has more smoking guns than a Texas shooting range, but this one is the smoking bazooka – a sign of criminal intent. It's unclear how most of the personal loans were used, but liquidators will have to recoup the money.

The loans to Paper Bird were even more worrisome because they created another related third party to shuffle assets. Forbes speculates that some Paper Bird funds went to buy Binance's FTX stake, and Paper Bird committed hundreds of millions to outside investments.

FTX Inner Circle: Who's Who

That included many FTX-backed VC funds. Time will tell if this financial incest was criminal fraud. It fits Bankman-pattern Fried's of using secret flows, leverage, and funny money to inflate asset prices.

FTT or loan 'bailouts'

Also. As the crypto bear market continued in 2022, Bankman-Fried proposed bailouts for bankrupt crypto lenders BlockFi and Voyager Digital. CoinDesk was among those deceived, welcoming SBF as a J.P. Morgan-style sector backstop.

In a now-infamous interview with CNBC's "Squawk Box," Bankman-Fried referred to these decisions as bets that may or may not pay off.

But maybe not. Bloomberg's Matt Levine speculated that FTX backed BlockFi with FTT money. This Monopoly bailout may have been intended to hide FTX and Alameda liabilities that would have been exposed if BlockFi went bankrupt sooner. This ploy has no name, but it echoes other corporate frauds.

Secret bank purchase

Alameda Research invested $11.5 million in the tiny Farmington State Bank, doubling its net worth. As a non-U.S. entity and an investment firm, Alameda should have cleared regulatory hurdles before acquiring a U.S. bank.

In the context of FTX, the bank's stake becomes "ominous." Alameda and FTX could have done more shenanigans with bank control. Compare this to the Bank for Credit and Commerce International's failed attempts to buy U.S. banks. BCCI was even nefarious than FTX and wanted to buy U.S. banks to expand its money-laundering empire.

The mainstream's mistakes

These are complex and nuanced forms of fraud that echo traditional finance models. This obscurity helped Bankman-Fried masquerade as an honest player and likely kept coverage soft after the collapse.

Bankman-Fried had a scruffy, nerdy image, like Mark Zuckerberg and Adam Neumann. In interviews, he spoke nonsense about an industry full of jargon and complicated tech. Strategic donations and insincere ideological statements helped him gain political and social influence.

SBF' s'Effective' Altruism Blew Up FTX

Bankman-Fried has continued to muddy the waters with disingenuous letters, statements, interviews, and tweets since his con collapsed. He's tried to portray himself as a well-intentioned but naive kid who made some mistakes. This is a softer, more pernicious version of what Trump learned from mob lawyer Roy Cohn. Bankman-Fried doesn't "deny, deny, deny" but "confuse, evade, distort."

It's mostly worked. Kevin O'Leary, who plays an investor on "Shark Tank," repeats Bankman-SBF's counterfactuals.  O'Leary called Bankman-Fried a "savant" and "probably one of the most accomplished crypto traders in the world" in a Nov. 27 interview with Business Insider, despite recent data indicating immense trading losses even when times were good.

O'Leary's status as an FTX investor and former paid spokesperson explains his continued affection for Bankman-Fried despite contradictory evidence. He's not the only one promoting Bankman-Fried. The disgraced son of two Stanford law professors will defend himself at Wednesday's DealBook Summit.

SBF's fraud and theft rival those of Bernie Madoff and Jho Low. Whether intentionally or through malign ineptitude, the fraud echoes Worldcom and Enron.

The Perverse Impacts of Anti-Money-Laundering

The principals in all of those scandals wound up either sentenced to prison or on the run from the law. Sam Bankman-Fried clearly deserves to share their fate.

Read the full article here.

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.