We've fought for humanity's sake. We need equilibrium.

We live in a world of opposites (black/white, up/down, love/hate), thus life is a game of achieving equilibrium. We have a universe of paradoxes within ourselves, not just in physics.

Individually, you balance your intellect and heart, but as a species, we're full of polarities. They might be gentle and compassionate, then ruthless and unsympathetic.

We desire for connection so much that we personify non-human beings and objects while turning to violence and hatred toward others. These contrasts baffle me. Will we find balance?

Anthropomorphization

Assigning human-like features or bonding with objects is common throughout childhood. Cartoons often give non-humans human traits. Adults still anthropomorphize this trait. Researchers agree we start doing it as infants and continue throughout life.

Humans of all ages are good at humanizing stuff. We build emotional attachments to weather events, inanimate objects, animals, plants, and locales. Gods, goddesses, and fictitious figures are anthropomorphized.

Cast Away, starring Tom Hanks, features anthropization. Hanks is left on an island, where he builds an emotional bond with a volleyball he calls Wilson.

We became emotionally invested in Wilson, including myself.

Why do we do it, though?

Our instincts and traits helped us survive and thrive. Our brain is alert to other people's thoughts, feelings, and intentions to assist us to determine who is safe or hazardous. We can think about others and our own mental states, or about thinking. This is the Theory of Mind.

Neurologically, specialists believe the Theory of Mind has to do with our mirror neurons, which exhibit the same activity while executing or witnessing an action.

Mirror neurons may contribute to anthropization, but they're not the only ones. In 2021, Harvard Medical School researchers at MGH and MIT colleagues published a study on the brain's notion of mind.

“Our study provides evidence to support theory of mind by individual neurons. Until now, it wasn’t clear whether or how neurons were able to perform these social cognitive computations.”

Neurons have particular functions, researchers found. Others encode information that differentiates one person's beliefs from another's. Some neurons reflect tale pieces, whereas others aren't directly involved in social reasoning but may multitask contributing factors.

Combining neuronal data gives a precise portrait of another's beliefs and comprehension. The theory of mind describes how we judge and understand each other in our species, and it likely led to anthropomorphism. Neuroscience indicates identical brain regions react to human or non-human behavior, like mirror neurons.

Some academics believe we're wired for connection, which explains why we anthropomorphize. When we're alone, we may anthropomorphize non-humans.

Humanizing non-human entities may make them deserving of moral care, according to another theory. Animamorphizing something makes it responsible for its actions and deserves punishments or rewards. This mental shift is typically apparent in our connections with pets and leads to deanthropomorphization.

Dehumanization

Dehumanizing involves denying someone or anything ethical regard, the opposite of anthropomorphizing.

Dehumanization occurs throughout history. We do it to everything in nature, including ourselves. We experiment on and torture animals. We enslave, hate, and harm other groups of people.

Race, immigrant status, dress choices, sexual orientation, social class, religion, gender, politics, need I go on? Our degrading behavior is promoting fascism and division everywhere.

Dehumanizing someone or anything reduces their agency and value. Many assume they're immune to this feature, but tests disagree.

It's inevitable. Humans are wired to have knee-jerk reactions to differences. We are programmed to dehumanize others, and it's easier than we'd like to admit.

Why do we do it, though?

Dehumanizing others is simpler than humanizing things for several reasons. First, we consider everything unusual as harmful, which has helped our species survive for hundreds of millions of years. Our propensity to be distrustful of others, like our fear of the unknown, promotes an us-vs.-them mentality.

Since WWII, various studies have been done to explain how or why the holocaust happened. How did so many individuals become radicalized to commit such awful actions and feel morally justified? Researchers quickly showed how easily the mind can turn gloomy.

Stanley Milgram's 1960s electroshock experiment highlighted how quickly people bow to authority to injure others. Philip Zimbardo's 1971 Stanford Prison Experiment revealed how power may be abused.

The us-versus-them attitude is natural and even young toddlers act on it. Without a relationship, empathy is more difficult.

It's terrifying how quickly dehumanizing behavior becomes commonplace. The current pandemic is an example. Most countries no longer count deaths. Long Covid is a major issue, with predictions of a handicapped tsunami in the future years. Mostly, we shrug.

In 2020, we panicked. Remember everyone's caution? Now Long Covid is ruining more lives, threatening to disable an insane amount of our population for months or their entire lives.

There's little research. Experts can't even classify or cure it. The people should be outraged, but most have ceased caring. They're over covid.

We're encouraged to find a method to live with a terrible pandemic that will cause years of damage. People aren't worried about infection anymore. They shrug and say, "We'll all get it eventually," then hope they're not one of the 30% who develops Long Covid.

We can correct course before further damage. Because we can recognize our urges and biases, we're not captives to them. We can think critically about our thoughts and behaviors, then attempt to improve. We can recognize our deficiencies and work to attain balance.

Changing perspectives

We're currently attempting to find equilibrium between opposites. It's superficial to defend extremes by stating we're only human or wired this way because both imply we have no control.

Being human involves having self-awareness, and by being careful of our thoughts and acts, we can find balance and recognize opposites' purpose.

Extreme anthropomorphizing and dehumanizing isolate and imperil us. We anthropomorphize because we desire connection and dehumanize because we're terrified, frequently of the connection we crave. Will we find balance?

Katrina Paulson ponders humanity, unanswered questions, and discoveries. Please check out her newsletters, Curious Adventure and Curious Life.

“I think if I had time to do something else, I would be so excited to go after this company right now.”

Sam Altman, CEO of Open AI, recently discussed AI's present and future.

Open AI is important. They're creating the cyberpunk and sci-fi worlds.

They use the most advanced algorithms and data sets.

GPT-3...sound familiar? Open AI built most copyrighting software. Peppertype, Jasper AI, Rytr. If you've used any, you'll be shocked by the quality.

Open AI isn't only GPT-3. They created DallE-2 and Whisper (a speech recognition software released last week).

What will they do next? What's the next great chance?

Sam Altman, CEO of Open AI, recently gave a lecture about the next trillion-dollar AI opportunity.

Who is the organization behind Open AI?

Open AI first. If you know, skip it.

Open AI is one of the earliest private AI startups. Elon Musk, Greg Brockman, and Rebekah Mercer established OpenAI in December 2015.

OpenAI has helped its citizens and AI since its birth.

They have scary-good algorithms.

Their GPT-3 natural language processing program is excellent.

The algorithm's exponential growth is astounding. GPT-2 came out in November 2019. May 2020 brought GPT-3.

Massive computation and datasets improved the technique in just a year. New York Times said GPT-3 could write like a human.

Same for Dall-E. Dall-E 2 was announced in April 2022. Dall-E 2 won a Colorado art contest.

Open AI's algorithms challenge jobs we thought required human innovation.

So what does Sam Altman think?

The Present Situation and AI's Limitations

During the interview, Sam states that we are still at the tip of the iceberg.

So I think so far, we’ve been in the realm where you can do an incredible copywriting business or you can do an education service or whatever. But I don’t think we’ve yet seen the people go after the trillion dollar take on Google.

He's right that AI can't generate net new human knowledge. It can train and synthesize vast amounts of knowledge, but it simply reproduces human work.

“It’s not going to cure cancer. It’s not going to add to the sum total of human scientific knowledge.”

But the key word is yet.

And that is what I think will turn out to be wrong that most surprises the current experts in the field.

Reinforcing his point that massive innovations are yet to come.

But where?

The Next $1 Trillion AI Company

Sam predicts a bio or genomic breakthrough.

There’s been some promising work in genomics, but stuff on a bench top hasn’t really impacted it. I think that’s going to change. And I think this is one of these areas where there will be these new $100 billion to $1 trillion companies started, and those areas are rare.

Avoid human trials since they take time. Bio-materials or simulators are suitable beginning points.

AI may have a breakthrough. DeepMind, an OpenAI competitor, has developed AlphaFold to predict protein 3D structures.

It could change how we see proteins and their function. AlphaFold could provide fresh understanding into how proteins work and diseases originate by revealing their structure. This could lead to Alzheimer's and cancer treatments. AlphaFold could speed up medication development by revealing how proteins interact with medicines.

Deep Mind offered 200 million protein structures for scientists to download (including sustainability, food insecurity, and neglected diseases).

Being in AI for 4+ years, I'm amazed at the progress. We're past the hype cycle, as evidenced by the collapse of AI startups like C3 AI, and have entered a productive phase.

We'll see innovative enterprises that could replace Google and other trillion-dollar companies.

What happens after AI adoption is scary and unpredictable. How will AGI (Artificial General Intelligence) affect us? Highly autonomous systems that exceed humans at valuable work (Open AI)

My guess is that the things that we’ll have to figure out are how we think about fairly distributing wealth, access to AGI systems, which will be the commodity of the realm, and governance, how we collectively decide what they can do, what they don’t do, things like that. And I think figuring out the answer to those questions is going to just be huge. — Sam Altman CEO

NASA's Artemis 1 mission will help return people to the Moon after a half-century break. The mission is a shakedown cruise for NASA's Space Launch System and Orion Crew Capsule.

The spaceship will visit the Moon, deploy satellites, and enter orbit. NASA wants to practice operating the spacecraft, test the conditions people will face on the Moon, and ensure a safe return to Earth.

We asked Jack Burns, a space scientist at the University of Colorado Boulder and former member of NASA's Presidential Transition Team, to describe the mission, explain what the Artemis program promises for space exploration, and reflect on how the space program has changed in the half-century since humans last set foot on the moon.

What distinguishes Artemis 1 from other rockets?

Artemis 1 is the Space Launch System's first launch. NASA calls this a "heavy-lift" vehicle. It will be more powerful than Apollo's Saturn V, which transported people to the Moon in the 1960s and 1970s.

It's a new sort of rocket system with two strap-on solid rocket boosters from the space shuttle. It's a mix of the shuttle and Saturn V.

The Orion Crew Capsule will be tested extensively. It'll spend a month in the high-radiation Moon environment. It will also test the heat shield, which protects the capsule and its occupants at 25,000 mph. The heat shield must work well because this is the fastest capsule descent since Apollo.

This mission will also carry miniature Moon-orbiting satellites. These will undertake vital precursor science, including as examining further into permanently shadowed craters where scientists suspect there is water and measuring the radiation environment to see long-term human consequences.

Artemis 1 will launch, fly to the Moon, place satellites, orbit it, return to Earth, and splash down in the ocean. NASA.

What's Artemis's goal? What launches are next?

The mission is a first step toward Artemis 3, which will lead to the first human Moon missions since 1972. Artemis 1 is unmanned.

Artemis 2 will have astronauts a few years later. Like Apollo 8, it will be an orbital mission that circles the Moon and returns. The astronauts will orbit the Moon longer and test everything with a crew.

Eventually, Artemis 3 will meet with the SpaceX Starship on the Moon's surface and transfer people. Orion will stay in orbit while the lunar Starship lands astronauts. They'll go to the Moon's south pole to investigate the water ice there.

Artemis is reminiscent of Apollo. What's changed in 50 years?

Kennedy wanted to beat the Soviets to the Moon with Apollo. The administration didn't care much about space flight or the Moon, but the goal would place America first in space and technology.

You live and die by the sword if you do that. When the U.S. reached the Moon, it was over. Russia lost. We planted flags and did science experiments. Richard Nixon canceled the program after Apollo 11 because the political goals were attained.

Large rocket with two boosters between two gates

NASA's new Space Launch System is brought to a launchpad. NASA

50 years later... It's quite different. We're not trying to beat the Russians, Chinese, or anyone else, but to begin sustainable space exploration.

Artemis has many goals. It includes harnessing in-situ resources like water ice and lunar soil to make food, fuel, and building materials.

SpaceX is part of this first journey to the Moon's surface, therefore the initiative is also helping to develop a lunar and space economy. NASA doesn't own the Starship but is buying seats for astronauts. SpaceX will employ Starship to transport cargo, private astronauts, and foreign astronauts.

Fifty years of technology advancement has made getting to the Moon cheaper and more practical, and computer technology allows for more advanced tests. 50 years of technological progress have changed everything. Anyone with enough money can send a spacecraft to the Moon, but not humans.

Commercial Lunar Payload Services engages commercial companies to develop uncrewed Moon landers. We're sending a radio telescope to the Moon in January. Even 10 years ago, that was impossible.

Since humans last visited the Moon 50 years ago, technology has improved greatly.

What other changes does Artemis have in store?

The government says Artemis 3 will have at least one woman and likely a person of color. 

I'm looking forward to seeing more diversity so young kids can say, "Hey, there's an astronaut that looks like me. I can do this. I can be part of the space program.”

These four, when combined, destroy procrastination.

You're not rushed. You waste it on busywork.

You'll accept this eventually.

• In 2022, the daily average usage of a user on social media is 2.5 hours.

• By 2020, 6 billion hours of video were watched each month by Netflix's customers, who used the service an average of 3.2 hours per day.

When we see these numbers, we think "Wow!" People squander so much time as though they don't contribute. True. These are yours. Likewise.

We don't lack time; we just waste it. Once you realize this, you can change your habits to save time. This article explains. If you adopt ALL 4 of these simple behaviors, you'll see amazing benefits.

Time-blocking

Cal Newport's time-blocking trick takes a minute but improves your day's clarity.

Divide the next day into 30-minute (or 5-minute, if you're Elon Musk) segments and assign responsibilities. As seen.

Here's why:

• The procrastination that results from attempting to determine when to begin working is eliminated. Procrastination is a given if you choose when to begin working in real-time. Even if you may assume you'll start working in five minutes, it won't take you long to realize that five minutes have turned into an hour. But if you've already determined to start working at 2:00 the next day, your odds of procrastinating are greatly decreased, if not eliminated altogether.

• You'll also see that you have a lot of time in a day when you plan your day out on paper and assign chores to each hour. Doing this daily will permanently eliminate the lack of time mindset.

5-4-3-2-1: Have breakfast with the frog!

“If it’s your job to eat a frog, it’s best to do it first thing in the morning. And If it’s your job to eat two frogs, it’s best to eat the biggest one first.”

Eating the frog means accomplishing the day's most difficult chore. It's better to schedule it first thing in the morning when time-blocking the night before. Why?

• The day's most difficult task is also the one that causes the most postponement. Because of the stress it causes, the later you schedule it, the more time you risk wasting by procrastinating.

• However, if you do it right away in the morning, you'll feel good all day. This is the reason it was set for the morning.

Mel Robbins' 5-second rule can help. Start counting backward 54321 and force yourself to start at 1. If you acquire the urge to work on a goal, you must act within 5 seconds or your brain will destroy it. If you're scheduled to eat your frog at 9, eat it at 8:59. Start working.

Micro-visualisation

You've heard of visualizing to enhance the future. Visualizing a bright future won't do much if you're not prepared to focus on the now and develop the necessary habits. Alexander said:

People don’t decide their futures. They decide their habits and their habits decide their future.

I visualize the next day's schedule every morning. My day looks like this

“I’ll start writing an article at 7:30 AM. Then, I’ll get dressed up and reach the medicine outpatient department by 9:30 AM. After my duty is over, I’ll have lunch at 2 PM, followed by a nap at 3 PM. Then, I’ll go to the gym at 4…”

etc.

This reinforces the day you planned the night before. This makes following your plan easy.

Set the timer.

It's the best iPhone productivity app. A timer is incredible for increasing productivity.

Set a timer for an hour or 40 minutes before starting work. Your call. I don't believe in techniques like the Pomodoro because I can focus for varied amounts of time depending on the time of day, how fatigued I am, and how cognitively demanding the activity is.

I work with a timer. A timer keeps you focused and prevents distractions. Your mind stays concentrated because of the timer. Timers generate accountability.

To pee, I'll pause my timer. When I sit down, I'll continue. Same goes for bottle refills. To use Twitter, I must pause the timer. This creates accountability and focuses work.

Connecting everything

If you do all 4, you won't be disappointed. Here's how:

• Plan out your day's schedule the night before.

• Next, envision in your mind's eye the same timetable in the morning.

• Speak aloud 54321 when it's time to work: Eat the frog! In the morning, devour the largest frog.

• Then set a timer to ensure that you remain focused on the task at hand.

Do you want to build and launch your own software company? To do this, all you need is a product that solves a problem.

Coming up with profitable ideas is not that easy. But you’re in luck because you got me!

I’ll give you the idea for free. All you need to do is execute it properly.

If you’re ready, let’s jump right into it! Starting with the problem.

Problem

Youtube has many creators. Every day, they think of new ways to entertain or inform us.

They work hard to make videos. Many of their efforts go to waste. They limit their revenue and reach.

Solution

Content repurposing solves this problem.

One video can become several TikToks. Creating YouTube videos from a podcast episode.

Or, one video might become a blog entry.

By turning videos into blog entries, Youtubers may develop evergreen SEO content, attract a new audience, and reach a non-YouTube audience.

Many YouTube creators want this easy feature.

Let's build it!

Implementation

We identified the problem, and we have a solution. All that’s left to do is see how it can be done.

Monitoring new video uploads

First, watch when a friend uploads a new video. Everything should happen automatically without user input.

YouTube Webhooks make this easy. Our server listens for YouTube Webhook notifications.

After publishing a new video, we create a conversion job.

Creating a Blog Post from a Video

Next, turn a video into a blog article.

To convert, we must extract the video's audio (which can be achieved by using FFmpeg on the server).

Once we have the audio channel, we can use speech-to-text.

Services can accomplish this easily.

• Speech-to-text on Google

• Google Translate

• Deepgram

Deepgram's affordability and integration make it my pick.

After conversion, the blog post needs formatting, error checking, and proofreading.

After this, a new blog post will appear in our web app's dashboard.

Completing a blog post

After conversion, users must examine and amend their blog posts.

Our application dashboard would handle all of this. It's a dashboard-style software where users can:

• Link their Youtube account

• Check out the converted videos in the future.

• View the conversions that are ongoing.

• Edit and format converted blog articles.

It's a web-based app.

It doesn't matter how it's made but I'd choose Next.js.

Next.js is a React front-end standard. Vercel serverless functions could conduct the conversions.

This would let me host the software for free and reduce server expenditures.

Taking It One Step Further

SaaS in a nutshell. Future improvements include integrating with WordPress or Ghost.

Our app users could then publish blog posts. Streamlining the procedure.

MVPs don't need this functionality.

Final Thoughts

Repurposing content helps you post more often, reach more people, and develop faster.

Many agencies charge a fortune for this service. Handmade means pricey.

Content creators will go crazy if you automate and cheaply solve this problem.

Just execute this idea!

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.