New social media producers' biggest error

The Treadmill of Social Media Content

New creators focus on the wrong platforms.

They post to Instagram, Twitter, TikTok, etc.

They create daily material, but it's never enough for social media algorithms.

Creators recognize they're on a content creation treadmill.

They have to keep publishing content daily just to stay on the algorithm’s good side and avoid losing the audience they’ve built on the platform.

This is exhausting and unsustainable, causing creator burnout.

They focus on short-lived platforms, which is an issue.

Comparing low- and high-return social media platforms

Social media networks are great for reaching new audiences.

Their algorithm is meant to viralize material.

Social media can use you for their aims if you're not careful.

To master social media, focus on the right platforms.

To do this, we must differentiate low-ROI and high-ROI platforms:

Low ROI platforms are ones where content has a short lifespan. High ROI platforms are ones where content has a longer lifespan.

A tweet may be shown for 12 days. If you write an article or blog post, it could get visitors for 23 years.

ROI is drastically different.

New creators have limited time and high learning curves.

Nothing is possible.

First create content for high-return platforms.

ROI for social media platforms

Here are high-return platforms:

1. Your Blog - A single blog article can rank and attract a ton of targeted traffic for a very long time thanks to the power of SEO.

2. YouTube - YouTube has a reputation for showing search results or sidebar recommendations for videos uploaded 23 years ago. A superb video you make may receive views for a number of years.

3. Medium - A platform dedicated to excellent writing is called Medium. When you write an article about a subject that never goes out of style, you're building a digital asset that can drive visitors indefinitely.

These high ROI platforms let you generate content once and get visitors for years.

This contrasts with low ROI platforms:

1. Twitter

2. Instagram

3. TikTok

4. LinkedIn

5. Facebook

The posts you publish on these networks have a 23-day lifetime. Instagram Reels and TikToks are exceptions since viral content can last months.

If you want to make content creation sustainable and enjoyable, you must focus the majority of your efforts on creating high ROI content first. You can then use the magic of repurposing content to publish content to the lower ROI platforms to increase your reach and exposure.

How To Use Your Content Again

So, you’ve decided to focus on the high ROI platforms.

Great!

You've published an article or a YouTube video.

You worked hard on it.

Now you have fresh stuff.

What now?

If you are not repurposing each piece of content for multiple platforms, you are throwing away your time and efforts.

You've created fantastic material, so why not distribute it across platforms?

Repurposing Content Step-by-Step

For me, it's writing a blog article, but you might start with a video or podcast.

The premise is the same regardless of the medium.

Start by creating content for a high ROI platform (YouTube, Blog Post, Medium). Then, repurpose, edit, and repost it to the lower ROI platforms.

Here's how to repurpose pillar material for other platforms:

1. Post the article on your blog.

2. Put your piece on Medium (use the canonical link to point to your blog as the source for SEO)

3. Create a video and upload it to YouTube using the talking points from the article.

4. Rewrite the piece a little, then post it to LinkedIn.

5. Change the article's format to a Thread and share it on Twitter.

6. Find a few quick quotes throughout the article, then use them in tweets or Instagram quote posts.

7. Create a carousel for Instagram and LinkedIn using screenshots from the Twitter Thread.

8. Go through your film and select a few valuable 30-second segments. Share them on LinkedIn, Facebook, Twitter, TikTok, YouTube Shorts, and Instagram Reels.

9. Your video's audio can be taken out and uploaded as a podcast episode.

If you (or your team) achieve all this, you'll have 20-30 pieces of social media content.

If you're just starting, I wouldn't advocate doing all of this at once.

Instead, focus on a few platforms with this method.

You can outsource this as your company expands. (If you'd want to learn more about content repurposing, contact me.)

You may focus on relevant work while someone else grows your social media on autopilot.

You develop high-ROI pillar content, and it's automatically chopped up and posted on social media.

This lets you use social media algorithms without getting sucked in.

Thanks for reading!

"Competition" is beautiful and hateful.

Some people bury their dreams because they are afraid of competition. Others challenge themselves, shaping our world.

Competition is normal.

It spurs innovation and progress.

I wish more people agreed.

As a marketer, content writer, and solopreneur, my readers often ask:

"I want to create a niche website, but I have no ideas. Everything's done"

"Is a website worthwhile?"

I can't count how many times I said, "Yes, it makes sense, and you can succeed in a competitive market."

I encourage and share examples, but it's not enough to overcome competition anxiety.

I launched an SEO writing website for content creators a year ago, knowing it wouldn't beat Ahrefs, Semrush, Backlinko, etc.

Not needed.

Many of my website's pages rank highly on Google.

Everyone can eat the pie.

In a competitive niche, I took a different approach.

Look farther

When chatting with bloggers that want a website, I discovered something fascinating.

They want to launch a website but have no ideas. As a next step, they start listing the interests they believe they should work on, like wellness, lifestyle, investments, etc. I could keep going.

Too many generalists who claim to know everything confuse many.

Generalists aren't trusted.

We want someone to fix our problems immediately.

I don't think broad-spectrum experts are undervalued. People have many demands that go beyond generalists' work. Narrow-niche experts can help.

I've done SEO for three years. I learned from experts and courses. I couldn't find a comprehensive SEO writing resource.

I read tons of articles before realizing that wasn't it. I took courses that covered SEO basics eventually.

I had a demand for learning SEO writing, but there was no solution on the market. My website fills this micro-niche.

Have you ever had trouble online?

Professional courses too general, boring, etc.?

You've bought off-topic books, right?

You're not alone.

Niche ideas!

Big players often disregard new opportunities. Too small. Individual content creators can succeed here.

In a competitive market:

• Never choose wide subjects

• Think about issues you can relate to and have direct experience with.

• Be a consumer to discover both the positive and negative aspects of a good or service.

• Merchandise your annoyances.

• Consider ways to transform your frustrations into opportunities.

The right niche is half-success. Here is what else I did to hit the Google front page with my website.

An innovative method for choosing subjects

Why publish on social media and websites?

Want likes, shares, followers, or fame?

Some people do it for fun. No judgment.

I bet you want more.

You want to make decent money from blogging.

Writing about random topics, even if they are related to your niche, won’t help you attract an audience from organic search. I'm a marketer and writer.

I worked at companies with dead blogs because they posted for themselves, not readers. They did not follow SEO writing rules; that’s why most of their content flopped.

I learned these hard lessons and grew my website from 0 to 3,000+ visitors per month while working on it a few hours a week only. Evidence:

I choose website topics using these criteria:

- Business potential. The information should benefit my audience and generate revenue. There would be no use in having it otherwise.

My topics should help me:

Attract organic search traffic with my "fluff-free" content -> Subscribers > SEO ebook sales.

Simple and effective.

- traffic on search engines. The number of monthly searches reveals how popular my topic is all across the world. If I find that no one is interested in my suggested topic, I don't write a blog article.

- Competition. Every search term is up against rivals. Some are more popular (thus competitive) since more websites target them in organic search. A new website won't score highly for keywords that are too competitive. On the other side, keywords with moderate to light competition can help you rank higher on Google more quickly.

- Search purpose. The "why" underlying users' search requests is revealed. I analyze search intent to understand what users need when they plug various queries in the search bar and what content can perfectly meet their needs.

My specialty website produces money, ranks well, and attracts the target audience because I handpick high-traffic themes.

Following these guidelines, even a new website can stand out.

I wrote a 50-page SEO writing guide where I detailed topic selection and share my front-page Google strategy.

My guide can help you run a successful niche website.

In summary

You're not late to the niche-website party.

The Internet offers many untapped opportunities.

We need new solutions and are willing to listen.

There are unexplored niches in any topic.

Don't fight giants. They have their piece of the pie. They might overlook new opportunities while trying to keep that piece of the pie. You should act now.

In 8 months, I doubled my audience with them.

LinkedIn's buzz isn't over.

People dream of social proof every day. They want clients, interesting jobs, and field recognition.

LinkedIn coaches will benefit greatly. Sell learning? Probably. Can you use it?

Consistency has been key in my eight-month study of LinkedIn. However, I'll share seven of my tips. 700 to 4500 people followed me.

1. Communication, communication, communication

LinkedIn is a social network. I like to think of it as a cafe. Here, you can share your thoughts, meet friends, and discuss life and work.

Do not treat LinkedIn as if it were a board for your post-its.

More socializing improves relationships. It's about people, like any network.

Consider interactions. Three main areas:

• Respond to criticism left on your posts.

• Comment on other people's posts

• Start and maintain conversations through direct messages.

Engage people. You spend too much time on Facebook if you only read your wall. Keeping in touch and having meaningful conversations helps build your network.

Every day, start a new conversation to make new friends.

2. Stick with those you admire

Interact thoughtfully.

Choose your contacts. Build your tribe is a term. Respectful networking.

I only had past colleagues, family, and friends in my network at the start of this year. Not business-friendly. Since then, I've sought out people I admire or can learn from.

Finding a few will help you. As they connect you to their networks. Friendships can lead to clients.

Don't underestimate network power. Cafe-style. Meet people at each table. But avoid people who sell SEO, web redesign, VAs, mysterious job opportunities, etc.

3. Share eye-catching infographics

Daily infographics flood LinkedIn. Visuals are popular. Use Canva's free templates if you can't draw them.

Last week's:

It's a fun way to visualize your topic.

You can repost and comment on infographics. Involve your network. I prefer making my own because I build my brand around certain designs.

My friend posted infographics consistently for four months and grew his network to 30,000.

If you start, credit the authors. As you steal someone's work.

4. Invite some friends over.

LinkedIn alone can be lonely. Having a few friends who support your work daily will boost your growth.

I was lucky to be invited to a group of networkers. We share knowledge and advice.

Having a few regulars who can discuss your posts is helpful. It's artificial, but it works and engages others.

Consider who you'd support if they were in your shoes.

You can pay for an engagement group, but you risk supporting unrelated people with rubbish posts.

Help each other out.

5. Don't let your feed or algorithm divert you.

LinkedIn's algorithm is magical.

Which time is best? How fast do you need to comment? Which days are best?

Overemphasize algorithms. Consider the user. No need to worry about the best time.

Remember to spend time on LinkedIn actively. Not passively. That is what Facebook is for.

Surely someone would find a LinkedIn recipe. Don't beat the algorithm yet. Consider your audience.

6. The more personal, the better

Personalization isn't limited to selfies. Share your successes and failures.

The more personality you show, the better.

People relate to others, not theories or quotes. Why should they follow you? Everyone posts the same content?

Consider your friends. What's their appeal?

Because they show their work and identity. It's simple. Medium and Linkedin are your platforms. Find out what works.

You can copy others' hooks and structures. You decide how simple to make it, though.

7. Have fun with those who have various post structures.

I like writing, infographics, videos, and carousels. Because you can:

Repurpose your content!

Out of one blog post I make:

• Newsletter

• Infographics (positive and negative points of view)

• Carousel

• Personal stories

• Listicle

Create less but more variety. Since LinkedIn posts last 24 hours, you can rotate the same topics for weeks without anyone noticing.

Effective!

The final LI snippet to think about

LinkedIn is about consistency. Some say 15 minutes. If you're serious about networking, spend more time there.

The good news is that it is worth it. The bad news is that it takes time.

VeeFriends is one NFT project I'm sure will last.

I believe in blockchain technology and JPEGs, aka NFTs. NFTs aren't JPEGs. It's not as it seems.

Gary Vaynerchuk is leading the pack with his new NFT project VeeFriends, I wrote a year ago. I was spot-on. It's the most innovative project I've seen.

Since its minting in May 2021, it has given its holders enormous value, most notably the first edition of VeeCon, a multi-day superconference featuring iconic and emerging leaders in NFTs and Popular Culture. First-of-its-kind NFT-ticketed Web3 conference to build friendships, share ideas, and learn together.

VeeFriends holders got free VeeCon NFT tickets. Attendees heard iconic keynote speeches, innovative talks, panels, and Q&A sessions.

It was a unique conference that most of us, including me, are looking forward to in 2023. The lineup was epic, and it allowed many to network in new ways. Really memorable learning. Here are a couple of gratitude posts from the attendees.

VeeFriends Series 2

This article explains VeeFriends if you're still confused.

GaryVee's hand-drawn doodles have evolved into wonderful characters. The characters' poses and backgrounds bring the VeeFriends IP to life.

Yes, this is the second edition of VeeFriends, and at current prices, it's one of the best NFT opportunities in years. If you have the funds and risk appetite to invest in NFTs, VeeFriends Series 2 is worth every penny. Even if you can't invest, learn from their journey.

1. Art Is the Start

Many critics say VeeFriends artwork is below average and not by GaryVee. Art is often the key to future success.

Let's look at one of the first Mickey Mouse drawings. No one would have guessed that this would become one of the most beloved animated short film characters. In Walt Before Mickey, Walt Disney's original mouse Mortimer was less refined.

First came a mouse...

These sketches evolved into Steamboat Willie, Disney's first animated short film.

Fred Moore redesigned the character artwork into what we saw in cartoons as kids. Mickey Mouse's history is here.

Looking at how different cartoon characters have evolved and gained popularity over decades, I believe Series 2 characters like Self-Aware Hare, Kind Kudu, and Patient Pig can do the same.

GaryVee captures this journey on the blockchain and lets early supporters become part of history. Time will tell if it rivals Disney, Pokemon, or Star Wars. Gary has been vocal about this vision.

2. VeeFriends is Intellectual Property for the Coming Generations

Most of us grew up watching cartoons, playing with toys, cards, and video games. Our interactions with fictional characters and the stories we hear shape us.

GaryVee is slowly curating an experience for the next generation with animated videos, card games, merchandise, toys, and more.

VeeFriends UNO, a collaboration with Mattel Creations, features 17 VeeFriends characters.

VeeFriends and Zerocool recently released Trading Cards featuring all 268 Series 1 characters and 15 new ones. Another way to build VeeFriends' collectibles brand.

At Veecon, all the characters were collectible toys. Something will soon emerge.

Kids and adults alike enjoy the YouTube channel's animated shorts and VeeFriends Tunes. Here's a song by the holder's Optimistic Otter-loving daughter.

This VeeFriends story is only the beginning. I'm looking forward to animated short film series, coloring books, streetwear, candy, toys, physical collectibles, and other forms of VeeFriends IP.

3. Veefriends will always provide utilities

Smart contracts can be updated at any time and authenticated on a ledger.

VeeFriends Series 2 gives no promise of any utility whatsoever. GaryVee released no project roadmap. In the first few months after launch, many owners of specific characters or scenes received utilities.

Every benefit or perk you receive helps promote the VeeFriends brand.

Recent partnerships are listed below.

• MaryRuth's Multivitamin Gummies

• Productive Puffin holders from VeeFriends x Primitive

• Pickleball Scene & Clown Holders Only

Pickleball & Competitive Clown Exclusive experience, anteater multivitamin gummies, and Puffin x Primitive merch

Considering the price of NFTs, it may not seem like much. It's just the beginning; you never know what the future holds. No other NFT project offers such diverse, ongoing benefits.

4. Garyvee's team is ready

Gary Vaynerchuk's team and record are undisputed. He's a serial entrepreneur and the Chairman & CEO of VaynerX, which includes VaynerMedia, VaynerCommerce, One37pm, and The Sasha Group.

Gary founded VaynerSports, Resy, and Empathy Wines. He's a Candy Digital Board Member, VCR Group Co-Founder, ArtOfficial Co-Founder, and VeeFriends Creator & CEO. Gary was recently named one of Fortune's Top 50 NFT Influencers.

Gary Vayenerchuk aka GaryVee

Gary documents his daily life as a CEO on social media, which has 34 million followers and 272 million monthly views. GaryVee Audio Experience is a top podcast. He's a five-time New York Times best-seller and sought-after speaker.

Gary can observe consumer behavior to predict trends. He understood these trends early and pioneered them.

• 1997 — Realized e-potential commerce's and started winelibrary.com. In five years, he grew his father's wine business from $3M to $60M.

• 2006 — Realized content marketing's potential and started Wine Library on YouTube. TV

• 2009 — Estimated social media's potential (Web2) and invested in Facebook, Twitter, and Tumblr.

• 2014: Ethereum and Bitcoin investments

• 2021 — Believed in NFTs and Web3 enough to launch VeeFriends

GaryVee isn't all of VeeFriends. Andy Krainak, Dave DeRosa, Adam Ripps, Tyler Dowdle, and others work tirelessly to make VeeFriends a success.

GaryVee has said he'll let other businesses fail but not VeeFriends. We're just beginning his 40-year vision.

I have more confidence than ever in a company with a strong foundation and team.

5. Humans die, but characters live forever

What if GaryVee dies or can't work?

A writer's books can immortalize them. As long as their books exist, their words are immortal. Socrates, Hemingway, Aristotle, Twain, Fitzgerald, and others have become immortal.

Everyone knows Vincent Van Gogh's The Starry Night.

We all love reading and watching Peter Parker, Thor, or Jessica Jones. Their behavior inspires us. Stan Lee's message and stories live on despite his death.

GaryVee represents VeeFriends. Creating characters to communicate ensures that the message reaches even those who don't listen.

Gary wants his values and messages to be omnipresent in 268 characters. Messengers die, but their messages live on.

Gary envisions VeeFriends creating timeless stories and experiences. Ten years from now, maybe every kid will sing Patient Pig.

6. I love the intent.

Gary planned to create Workplace Warriors three years ago when he began designing Patient Panda, Accountable Ant, and Empathy elephant. The project stalled. When NFTs came along, he knew.

Gary wanted to create characters with traits he values, such as accountability, empathy, patience, kindness, and self-awareness. He wants future generations to find these traits cool. He hopes one or more of his characters will become pop culture icons.

These emotional skills aren't taught in schools or colleges, but they're crucial for business and life success. I love that someone is teaching this at scale.

In the end, intent matters.

Humans Are Collectors

Buy and collect things to communicate. Since the 1700s. Medieval people formed communities around hidden metals and stones. Many people still collect stamps and coins, and luxury and fashion are multi-trillion dollar industries. We're collectors.

The early 2020s NFTs will be remembered in the future. VeeFriends will define a cultural and technological shift in this era. VeeFriends Series 1 is the original hand-drawn art, but it's expensive. VeeFriends Series 2 is a once-in-a-lifetime opportunity at $1,000.

If you are new to NFTs, check out How to Buy a Non Fungible Token (NFT) For Beginners

This is a non-commercial article. Not financial or legal advice. Information isn't always accurate. Before making important financial decisions, consult a pro or do your own research.

This post is a summary. 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.

Russia's nuclear threat may be nullified by physics.

Putin seems nostalgic and wants to relive the Cold War. He's started a deadly war to reclaim the old Soviet state of Ukraine and is threatening the West with nuclear war. NATO can't risk starting a global nuclear war that could wipe out humanity to support Ukraine's independence as much as they want to. Fortunately, nuclear physics may have rendered Putin's nuclear weapons useless. However? How will Ukraine and NATO react?

To understand why Russia's nuclear weapons may be ineffective, we must first know what kind they are.

Russia has the world's largest nuclear arsenal, with 4,447 strategic and 1,912 tactical weapons (all of which are ready to be rolled out quickly). The difference between these two weapons is small, but it affects their use and logistics. Strategic nuclear weapons are ICBMs designed to destroy a city across the globe. Russia's ICBMs have many designs and a yield of 300–800 kilotonnes. 300 kilotonnes can destroy Washington. Tactical nuclear weapons are smaller and can be fired from artillery guns or small truck-mounted missile launchers, giving them a 1,500 km range. Instead of destroying a distant city, they are designed to eliminate specific positions, bases, or military infrastructure. They produce 1–50 kilotonnes.

These two nuclear weapons use different nuclear reactions. Pure fission bombs are compact enough to fit in a shell or small missile. All early nuclear weapons used this design for their fission bombs. This technology is inefficient for bombs over 50 kilotonnes. Larger bombs are thermonuclear. Thermonuclear weapons use a small fission bomb to compress and heat a hydrogen capsule, which undergoes fusion and releases far more energy than ignition fission reactions, allowing for effective giant bombs. 

Here's Russia's issue.

A thermonuclear bomb needs deuterium (hydrogen with one neutron) and tritium (hydrogen with two neutrons). Because these two isotopes fuse at lower energies than others, the bomb works. One problem. Tritium is highly radioactive, with a half-life of only 12.5 years, and must be artificially made.

Tritium is made by irradiating lithium in nuclear reactors and extracting the gas. Tritium is one of the most expensive materials ever made, at $30,000 per gram.

Why does this affect Putin's nukes?

Thermonuclear weapons need tritium. Tritium decays quickly, so they must be regularly refilled at great cost, which Russia may struggle to do.

Russia has a smaller economy than New York, yet they are running an invasion, fending off international sanctions, and refining tritium for 4,447 thermonuclear weapons.

The Russian military is underfunded. Because the state can't afford it, Russian troops must buy their own body armor. Arguably, Putin cares more about the Ukraine conflict than maintaining his nuclear deterrent. Putin will likely lose power if he loses the Ukraine war.

It's possible that Putin halted tritium production and refueling to save money for Ukraine. His threats of nuclear attacks and escalating nuclear war may be a bluff.

This doesn't help Ukraine, sadly. Russia's tactical nuclear weapons don't need expensive refueling and will help with the invasion. So Ukraine still risks a nuclear attack. The bomb that destroyed Hiroshima was 15 kilotonnes, and Russia's tactical Iskander-K nuclear missile has a 50-kiloton yield. Even "little" bombs are deadly.

We can't guarantee it's happening in Russia. Putin may prioritize tritium. He knows the power of nuclear deterrence. Russia may have enough tritium for this conflict. Stockpiling a material with a short shelf life is unlikely, though.

This means that Russia's most powerful weapons may be nearly useless, but they may still be deadly. If true, this could allow NATO to offer full support to Ukraine and push the Russian tyrant back where he belongs. If Putin withholds funds from his crumbling military to maintain his nuclear deterrent, he may be willing to sink the ship with him. Let's hope the former.