I recently wrote about getting VCs excited about Marketplace start-ups. SaaS founders became envious!

Understanding how people wire tens of millions is the only Series A hack I recommend.

Few people understand the intellectual process behind investing.

VC is risk management.

Series A-focused VCs must cover two risks.

1. Market risk

You need a large market to cross a threshold beyond which you can build defensibilities. Series A VCs underwrite market risk.

They must see you have reached product-market fit (PMF) in a large total addressable market (TAM).

2. Execution risk

When evaluating your growth engine's blitzscaling ability, execution risk arises.

When investors remove operational uncertainty, they profit.

Series A VCs like businesses with derisked revenue streams. Don't raise unless you have a predictable model, pipeline, and growth.

Please beat these 3 metrics before Series A:

Achieve $1.5m ARR in 12-24 months (Market risk)

Above 100% Net Dollar Retention. (Market danger)

Lead Velocity Rate supporting $10m ARR in 2–4 years (Execution risk)

Hit the 3 and you'll raise $10M in 4 months. Discussing 2/3 may take 6–7 months.

If none, don't bother raising and focus on becoming a capital-efficient business (Topics for other posts).

Let's examine these 3 metrics for the brave ones.

1. Lead Velocity Rate supporting €$10m ARR in 2 to 4 years

Last because it's the least discussed. LVR is the most reliable data when evaluating a growth engine, in my opinion.

SaaS allows you to see the future.

Monthly Sales and Sales Pipelines, two predictive KPIs, have poor data quality. Both are lagging indicators, and minor changes can cause huge modeling differences.

Analysts and Associates will trash your forecasts if they're based only on Monthly Sales and Sales Pipeline.

LVR, defined as month-over-month growth in qualified leads, is rock-solid. There's no lag. You can See The Future if you use Qualified Leads and a consistent formula and process to qualify them.

With this metric in your hand, scaling your company turns into an execution play on which VCs are able to perform calculations risk.

2. Above-100% Net Dollar Retention.

Net Dollar Retention is a better-known SaaS health metric than LVR.

Net Dollar Retention measures a SaaS company's ability to retain and upsell customers. Ask what $1 of net new customer spend will be worth in years n+1, n+2, etc.

Depending on the business model, SaaS businesses can increase their share of customers' wallets by increasing users, selling them more products in SaaS-enabled marketplaces, other add-ons, and renewing them at higher price tiers.

If a SaaS company's annualized Net Dollar Retention is less than 75%, there's a problem with the business.

Slack's ARR chart (below) shows how powerful Net Retention is. Layer chart shows how existing customer revenue grows. Slack's S1 shows 171% Net Dollar Retention for 2017–2019.

Slack S-1

3. $1.5m ARR in the last 12-24 months.

According to Point 9, $0.5m-4m in ARR is needed to raise a $5–12m Series A round.

Target at least what you raised in Pre-Seed/Seed. If you've raised $1.5m since launch, don't raise before $1.5m ARR.

Capital efficiency has returned since Covid19. After raising $2m since inception, it's harder to raise $1m in ARR.

P9's 2016-2021 SaaS Funding Napkin

In summary, less than 1% of companies VCs meet get funded. These metrics can help you win.

If there’s demand for it, I’ll do one on direct-to-consumer.

Cheers!

You do not want to miss these

Stop! Stop it! Two things to avoid when starting a YouTube channel. Critical. Possible channel-killers Its future revenue.

I'll tell you now, so don't say "I wish I knew."

The Notorious Copyright Allegation

My YouTube channel received a copyright claim before I sold it. This claim was on a one-minute video I thought I'd changed enough to make mine, but the original owner disagreed.

It cost me thousands in ad revenue. Original owner got the profits.

Well, it wasn't your video, you say.

Touché.

I've learned. Sorta

I couldn't stop looking at the video's views. The video got 1,000,000 views without any revenue. I made 4 more similar videos.

If they didn't get copyrighted, I'd be rolling in dough.

You've spent a week editing and are uploading to YouTube. You're thrilled as you stand and stretch your back. You see the video just before publishing.

No way!

The red exclamation point on checks.

Copyright claim!

YouTube lets you publish, but you won't make money.

Sounds fair? Well, it is.

Copyright claims mean you stole someone's work. Song, image, or video clip.

We wouldn't want our content used for money.

The only problem with this is that almost everything belongs to someone else. I doubt some of the biggest creators are sitting down and making their music for their videos. That just seems really excessive when you could make a quick search on YouTube and download a song (I definitely don’t do this because that would be stealing).

So how do you defeat a copyright defense?

Even copyright-free songs on YouTube aren't guaranteed. Some copyrighted songs claim to be free.

Use YouTube's free music library or pay for a subscription to adobe stock, epidemic sound, or artlist.io.

Most of my videos have Nintendo music. Almost all game soundtracks are copyright-free and offer a variety of songs.

Restriction on age

Age restrictions are a must-avoid. A channel dies.

YouTube never suggests age-restricted videos.

Shadow banning means YouTube hides your content from subscribers and non-subscribers.

Keeping your channel family-friendly can help.

I hear you complaining that your channel isn't for kids. I agree. Not everyone has a clean mouth or creates content for minors.

YouTube has changed rapidly in recent years. Focusing on kids. Fewer big creators are using profanity or explicit content in videos. Not YouTube-worthy.

Youtube wants to be family-friendly. A family-friendly movie. It won't promote illegal content. Yes, it allows profanity.

Do I recommend avoiding no-no words in videos? Never. Okay. YouTube's policies are shaky. YouTube uses video content to determine ad suitability.

No joke. If you're serious about becoming a content creator, avoid profanity and inappropriate topics.

If your channel covers 18+ topics, like crime or commentary, censor as much as possible.

YouTube can be like walking on eggshells. You never know what is gonna upset the boss. So play it safe and try to avoid getting on their bad side.

Mr. Beast, Dream, Markplier, Faze Rug, and PewDewPie are popular creators. They maintain it family-friendly while entertaining fans.

You got this.

The comeback tour continues…

First, he founded a $47 billion co-working company (sorry, a “tech company”).

He established WeLive to disrupt apartment life.

Then he created WeGrow, a school that tossed aside the usual curriculum to feed children's souls and release their potential.

He raised the world’s consciousness.

Then he blew it all up (without raising the world’s consciousness). (He bought a wave pool.)

Adam Neumann's WeWork business burned investors' money. The founder sailed off with unimaginable riches, leaving long-time employees with worthless stocks and the company bleeding money. His track record, which includes a failing baby clothing company, should have stopped investors cold.

Once the dust settled, folks went on. We forgot about the Neumanns! We forgot about the private jets, company retreats, many houses, and WeWork's crippling. In that moment, the prodigal son of entrepreneurship returned, choosing the blockchain as his industry. His homecoming tour began with Flowcarbon, which sold Goddess Nature Tokens to lessen companies' carbon footprints.

Did it work?

Of course not.

Despite receiving $70 million from Andreessen Horowitz's a16z, the project has been halted just two months after its announcement.

This triumph should lower his grade.

Neumann seems to have moved on and has another revolutionary idea for the future of living. Flow (not Flowcarbon) aims to help people live in flow and will launch in 2023. It's the classic Neumann pitch: lofty goals, yogababble, and charisma to attract investors.

It's a winning formula for one investment fund. a16z has backed the project with its largest single check, $350 million. It has a splash page and 3,000 rental units, but is valued at over $1 billion. The blog post praised Neumann for reimagining the office and leading a paradigm-shifting global company.

Flow's mission is to solve the nation's housing crisis. How? Idk. It involves offering community-centric services in apartment properties to the same remote workforce he once wooed with free beer and a pingpong table. Revolutionary! It seems the goal is to apply WeWork's goals of transforming physical spaces and building community to apartments to solve many of today's housing problems.

The elevator pitch probably sounded great.

At least a16z knows it's a near-impossible task, calling it a seismic shift. Marc Andreessen opposes affordable housing in his wealthy Silicon Valley town. As details of the project emerge, more investors will likely throw ethics and morals out the window to go with the flow, throwing money at a man known for burning through it while building toxic companies, hoping he can bank another fantasy valuation before it all crashes.

Insanity is repeating the same action and expecting a different result. Everyone on the Neumann hype train needs to sober up.

Like WeWork, this venture Won’tWork.

Like before, it'll cause a shitstorm.

Recent investigations have shown that generative AI can boost hackers and misinformation spreaders.

Since its inception in late November 2022, OpenAI's ChatGPT has entertained and assisted many online users in writing, coding, task automation, and linguistic translation. Given this versatility, it is maybe unsurprising but nonetheless regrettable that fraudsters and mis-, dis-, and malinformation (MDM) spreaders are also considering ChatGPT and related AI models to streamline and improve their operations.

Malign actors may benefit from ChatGPT, according to a WithSecure research. ChatGPT promises to elevate unlawful operations across many attack channels. ChatGPT can automate spear phishing attacks that deceive corporate victims into reading emails from trusted parties. Malware, extortion, and illicit fund transfers can result from such access.

ChatGPT's ability to simulate a desired writing style makes spear phishing emails look more genuine, especially for international actors who don't speak English (or other languages like Spanish and French).

This technique could let Russian, North Korean, and Iranian state-backed hackers conduct more convincing social engineering and election intervention in the US. ChatGPT can also create several campaigns and various phony online personas to promote them, making such attacks successful through volume or variation. Additionally, image-generating AI algorithms and other developing techniques can help these efforts deceive potential victims.

Hackers are discussing using ChatGPT to install malware and steal data, according to a Check Point research. Though ChatGPT's scripts are well-known in the cyber security business, they can assist amateur actors with little technical understanding into the field and possibly develop their hacking and social engineering skills through repeated use.

Additionally, ChatGPT's hacking suggestions may change. As a writer recently indicated, ChatGPT's ability to blend textual and code-based writing might be a game-changer, allowing the injection of innocent content that would subsequently turn out to be a malicious script into targeted systems. These new AI-powered writing- and code-generation abilities allow for unique cyber attacks, regardless of viability.

OpenAI fears ChatGPT usage. OpenAI, Georgetown University's Center for Security and Emerging Technology, and Stanford's Internet Observatory wrote a paper on how AI language models could enhance nation state-backed influence operations. As a last resort, the authors consider polluting the internet with radioactive or misleading data to ensure that AI language models produce outputs that other language models can identify as AI-generated. However, the authors of this paper seem unaware that their "solution" might cause much worse MDM difficulties.

Literally False News

The public argument about ChatGPTs content-generation has focused on originality, bias, and academic honesty, but broader global issues are at stake. ChatGPT can influence public opinion, troll individuals, and interfere in local and national elections by creating and automating enormous amounts of social media material for specified audiences.

ChatGPT's capacity to generate textual and code output is crucial. ChatGPT can write Python scripts for social media bots and give diverse content for repeated posts. The tool's sophistication makes it irrelevant to one's language skills, especially English, when writing MDM propaganda.

I ordered ChatGPT to write a news piece in the style of big US publications declaring that Ukraine is on the verge of defeat in its fight against Russia due to corruption, desertion, and exhaustion in its army. I also gave it a fake reporter's byline and an unidentified NATO source's remark. The outcome appears convincing:

Worse, terrible performers can modify this piece to make it more credible. They can edit the general's name or add facts about current wars. Furthermore, such actors can create many versions of this report in different forms and distribute them separately, boosting its impact.

In this example, ChatGPT produced a news story regarding (fictional) greater moviegoer fatality rates:

Editing this example makes it more plausible. Dr. Jane Smith, the putative author of the medical report, might be replaced with a real-life medical person or a real victim of this supposed medical hazard.

Can deceptive texts be found? Detecting AI text is behind AI advancements. Minor AI-generated text alterations can upset these technologies.

Some OpenAI individuals have proposed covert methods to watermark AI-generated literature to prevent its abuse. AI models would create information that appears normal to humans but would follow a cryptographic formula that would warn other machines that it was AI-made. However, security experts are cautious since manually altering the content interrupts machine and human detection of AI-generated material.

How to Prepare

Cyber security and IT workers can research and use generative AI models to fight spear fishing and extortion. Governments may also launch MDM-defence projects.

In election cycles and global crises, regular people may be the most vulnerable to AI-produced deceit. Until regulation or subsequent technical advances, individuals must recognize exposure to AI-generated fraud, dating scams, other MDM activities.

A three-step verification method of new material in suspicious emails or social media posts can help identify AI content and manipulation. This three-step approach asks about the information's distribution platform (is it reliable? ), author (is the reader familiar with them? ), and plausibility given one's prior knowledge of the topic.

Consider a report by a trusted journalist that makes shocking statements in their typical manner. AI-powered fake news may be released on an unexpected platform, such as a newly created Facebook profile. However, if it links to a known media source, it is more likely to be real.

Though hard and subjective, this verification method may be the only barrier against manipulation for now.

AI language models:

How to Recognize an AI-Generated Article ChatGPT, the popular AI-powered chatbot, can and likely does generate medium.com-style articles.

AI-Generated Text Detectors Fail. Do This. Online tools claim to detect ChatGPT output. Even with superior programming, I tested some of these tools. pub

Why Original Writers Matter Despite AI Language Models Creative writers may never be threatened by AI language models.

Apple's latest revelation may shock iPod fans and former owners.

Apple discontinued the iPod touch on May 11, 2022. After 21 years, Apple killed the last surviving iPod, a device Steve Jobs believed would revolutionize the music industry.

Jobs was used to making bold predictions, but few expected Apple's digital music player to change the music industry. It did.

This chaos created new business opportunities. Spotify, YouTube, and Amazon are products of that chaotic era.

As the digital landscape changes, so do consumers, and the iPod has lost favor. I'm sure Apple realizes the importance of removing an icon. The iPod was Apple like the Mac and iPhone. I think it's bold to retire such a key Apple cornerstone. What would Jobs do?

iPod evolution across the ages

Here's an iPod family tree for all you enthusiasts.

iPod vintage (Oct 2001 to Sep 2014, 6 generations)

The original iPod had six significant upgrades since 2001. Apple announced an 80 GB ($249) and 160 GB ($349) iPod classic in 2007.

Apple updated the 80 GB model with a 120 GB device in September 2008. Apple upgraded the 120 GB model with a 160 GB variant a year later (2009). This was the last iteration, and Apple discontinued the classic in September 2014.

iPod nano (Jan 2004 to Sep 2005, 2 generations)

Apple debuted a smaller, brightly-colored iPod in 2004. The first model featured 4 GB, enough for 1,000 songs.

Apple produced a new 4 GB or 6 GB iPod mini in February 2005 and discontinued it in September when they released a better-looking iPod nano.

iTouch nano (Sep 2005 to July 2017, 7 generations)

I loved the iPod nano. It was tiny and elegant with enough tech to please most music aficionados, unless you carry around your complete music collection.

Apple owed much of the iPod nano's small form and success to solid-state flash memory. Flash memory doesn't need power because it has no moving parts. This makes the iPod nano more durable than the iPod classic and mini, which employ hard drives.

Apple manufactured seven generations of the iPod nano, improving its design, display screen, memory, battery, and software, but abandoned it in July 2017 due to dwindling demand.

Shuffle iPod (Jan 2005 to Jul 2017, 4 generations)

The iPod shuffle was entry-level. It was a simple, lightweight, tiny music player. The iPod shuffle was perfect for lengthy bike trips, runs, and hikes.

Apple sold 10 million iPod shuffles in the first year and kept making them for 12 years, through four significant modifications.

iOS device (Sep 2007 to May 2022, 7 generations)

The iPod touch's bigger touchscreen interface made it a curious addition to the iPod family. The iPod touch resembled an iPhone more than the other iPods, making them hard to tell apart.

Many were dissatisfied that Apple removed functionality from the iPod touch to avoid making it too similar to the iPhone. Seven design improvements over 15 years brought the iPod touch closer to the iPhone, but not completely.

The iPod touch uses the same iOS operating system as the iPhone, giving it access to many apps, including handheld games.

The iPod touch's long production run is due to the next generation of music-loving gamers.

What made the iPod cool

iPod revolutionized music listening. It was the first device to store and play MP3 music, allowing you to carry over 1,000 songs anywhere.

The iPod changed consumer electronics with its scroll wheel and touchscreen. Jobs valued form and function equally. He showed people that a product must look good to inspire an emotional response and ignite passion.

The elegant, tiny iPod was a tremendous sensation when it arrived for $399 in October 2001. Even at this price, it became a must-have for teens to CEOs.

It's hard to identify any technology that changed how music was downloaded and played like the iPod. Apple iPod and iTunes had 63% of the paid music download market in the fourth quarter of 2012.

The demise of the iPod was inevitable

Apple discontinuing the iPod touch after 21 years is sad. This ends a 00s music icon.

Jobs was a genius at anticipating market needs and opportunities, and Apple launched the iPod at the correct time.

Few consumer electronics items have had such a lasting impact on music lovers and the music industry as the iPod.

Smartphones and social media have contributed to the iPod's decline. Instead of moving to the music, the new generation of consumers is focused on social media. They're no longer passive content consumers; they're active content creators seeking likes and followers. Here, the smartphone has replaced the iPod.

It's hard not to feel a feeling of loss, another part of my adolescence now forgotten by the following generation.

So, if you’re lucky enough to have a working iPod, hang on to that relic and enjoy the music and the nostalgia.

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.