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.

The trick Notion could use is simply Facebook-ing the hell out of them.

*Time travel to fifteen years ago.* Future-Me: “Hey! What are you up to?” Old-Me: “I am proofreading an article. It’s taking a few hours, but I will be done soon.” Future-Me: “You know, in the future, you will be using a google chrome plugin called Grammarly that will help you easily proofread articles in half that time.” Old-Me: “What is… Google Chrome?” Future-Me: “Gosh…”

I love Grammarly. It’s one of those products that I personally feel the effects of. I mean, Space X is a great company. But I am not a rocket writing this article in space (or am I?)…

No, I’m not. So I don’t personally feel a connection to Space X. So, if a company collapse occurs in the morning, I might write about it. But I will have zero emotions regarding it.

Yet, if Grammarly fails tomorrow, I will feel 1% emotionally distressed. So looking at the title of this article, you’d realize that I am betting against them. This is how much I believe in the critical business model that’s taking over the world, the one of Notion.

Notion How frequently do you go through your notes?

Grammarly is everywhere, which helps its success. Grammarly is available when you update LinkedIn on Chrome. Grammarly prevents errors in Google Docs.

My internal concentration isn't apparent in the previous paragraph. Not Grammarly. I should have used Chrome to make a Google doc and LinkedIn update. Without this base, Grammarly will be useless.

So, welcome to this business essay.

• Grammarly provides a solution.

• Another issue is resolved by Jasper.

• Your entire existence is supposed to be contained within Notion.

New Google Chrome is offline. It's an all-purpose notepad (in the near future.)

• How should I start my blog? Enter it in Note.

• an update on LinkedIn? If you mention it, it might be automatically uploaded there (with little help from another app.)

• An advanced thesis? You can brainstorm it with your coworkers.

This ad sounds great! I won't cry if Notion dies tomorrow.

I'll reread the following passages to illustrate why I think Notion could kill Grammarly and Jasper.

Notion is a fantastic app that incubates your work.

Smartly, they began with note-taking.

Hopefully, your work will be on Notion. Grammarly and Jasper are still must-haves.

Grammarly will proofread your typing while Jasper helps with copywriting and AI picture development.

They're the best, therefore you'll need them. Correct? Nah.

Notion might bombard them with Facebook posts.

Notion: “Hi Grammarly, do you want to sell your product to us?” Grammarly: “Dude, we are more valuable than you are. We’ve even raised $400m, while you raised $342m. Our last valuation round put us at $13 billion, while yours put you at $10 billion. Go to hell.” Notion: “Okay, we’ll speak again in five years.”

Notion: “Jasper, wanna sell?” Jasper: “Nah, we’re deep into AI and the field. You can’t compete with our people.” Notion: “How about you either sell or you turn into a Snapchat case?” Jasper: “…”

Notion is your home. Grammarly is your neighbor. Your track is Jasper.

What if you grew enough vegetables in your backyard to avoid the supermarket? No more visits.

What if your home had a beautiful treadmill? You won't rush outside as much (I disagree with my own metaphor). (You get it.)

It's Facebooking. Instagram Stories reduced your Snapchat usage. Notion will reduce your need to use Grammarly.

The Final Piece of the AI Puzzle

Let's talk about Notion first, since you've probably read about it everywhere.

• They raised $343 million, as I previously reported, and bought four businesses

• According to Forbes, Notion will have more than 20 million users by 2022. The number of users is up from 4 million in 2020.

If raising $1.8 billion was impressive, FTX wouldn't have fallen.

This article compares the basic product to two others. Notion is a day-long app.

Notion has released Notion AI to support writers. It's early, so it's not as good as Jasper. Then-Jasper isn't now-Jasper. In five years, Notion AI will be different.

With hard work, they may construct a Jasper-like writing assistant. They have resources and users.

At this point, it's all speculation. Jasper's copywriting is top-notch. Grammarly's proofreading is top-notch. Businesses are constrained by user activities.

If Notion's future business movements are strategic, they might become a blue ocean shark (or get acquired by an unbelievable amount.)

I love business mental teasers, so tell me:

• How do you feel? Are you a frequent Notion user?

• Do you dispute my position? I enjoy hearing opposing viewpoints.

Ironically, I proofread this with Grammarly.

Generative AI hype: some thoughts

The sudden surge in "generative AI" startups and projects feels like the inverse of the recent "web3" boom. Both came from hyped-up pots. But while web3 hyped idealistic tech and an easy way to make money, generative AI hypes unsettling tech and questions whether it can be used to make money.

Web3 is technology looking for problems to solve, while generative AI is technology creating almost too many solutions. Web3 has been evangelists trying to solve old problems with new technology. As Generative AI evolves, users are resolving old problems in stunning new ways.

It's a jab at web3, but it's true. Web3's hype, including crypto, was unhealthy. Always expected a tech crash and shakeout. Tech that won't look like "web3" but will enhance "web2"

But that doesn't mean AI hype is healthy. There'll be plenty of bullshit here, too. As moths to a flame, hype attracts charlatans. Again, the difference is the different starting point. People want to use it. Try it.

With the beta launch of Dall-E 2 earlier this year, a new class of consumer product took off. Midjourney followed suit (despite having to jump through the Discord server hoops). Twelve more generative art projects. Lensa, Prisma Labs' generative AI self-portrait project, may have topped the hype (a startup which has actually been going after this general space for quite a while). This week, ChatGPT went off-topic.

This has a "fake-it-till-you-make-it" vibe. We give these projects too much credit because they create easy illusions. This also unlocks new forms of creativity. And faith in new possibilities.

As a user, it's thrilling. We're just getting started. These projects are not only fun to play with, but each week brings a new breakthrough. As an investor, it's all happening so fast, with so much hype (and ethical and societal questions), that no one knows how it will turn out. Web3's demand won't be the issue. Too much demand may cause servers to melt down, sending costs soaring. Companies will try to mix rapidly evolving tech to meet user demand and create businesses. Frustratingly difficult.

Anyway, I wanted an excuse to post some Lensa selfies.

These are really weird. I recognize them as me or a version of me, but I have no memory of them being taken. It's surreal, out-of-body. Uncanny Valley.

My kids redesigned the internet because it lacks inventiveness.

Internet today is bland. Everything is generic: fonts, layouts, pages, and visual language. Microtypography is messy.

Web design today seems dictated by technical and ideological constraints rather than creativity and ideas. Text and graphics are in containers on every page. All design is assumed.

Ironically, web technologies can design a lot. We can execute most designs. We make shocking, evocative websites. Experimental typography, generating graphics, and interactive experiences are possible.

Even designer websites use containers in containers. Dribbble and Behance, the two most popular creative websites, are boring. Lead image.

How did this happen?

Several reasons. WordPress and other blogging platforms use templates. These frameworks build web pages by combining graphics, headlines, body content, and videos. Not designs, templates. These rules combine related data types. These platforms don't let users customize pages beyond the template. You filled the template.

Templates are content-neutral. Thus, the issue.

Form should reflect and shape content, which is a design principle. Separating them produces content containers. Templates have no design value.

One of the fundamental principles of design is a deep and meaningful connection between form and content.

Web design lacks imagination for many reasons. Most are pragmatic and economic. Page design takes time. Large websites lack the resources to create a page from scratch due to the speed of internet news and the frequency of new items. HTML, JavaScript, and CSS continue to challenge web designers. Web design can't match desktop publishing's straightforward operations.

Designers may also be lazy. Mobile-first, generic, framework-driven development tends to ignore web page visual and contextual integrity.

How can we overcome this? How might expressive and avant-garde websites look today?

Rediscovering the past helps design the future.

'90s-era web design

At the University of the Arts Bremen's research and development group, I created my first website 23 years ago. Web design was trendy. Young web. Pages inspired me.

We struggled with HTML in the mid-1990s. Arial, Times, and Verdana were the only web-safe fonts. Anything exciting required table layouts, monospaced fonts, or GIFs. HTML was originally content-driven, thus we had to work against it to create a page.

Experimental typography was booming. Designers challenged the established quo from Jan Tschichold's Die Neue Typographie in the twenties to April Greiman's computer-driven layouts in the eighties. By the mid-1990s, an uncommon confluence of technological and cultural breakthroughs enabled radical graphic design. Irma Boom, David Carson, Paula Scher, Neville Brody, and others showed it.

Early web pages were dull compared to graphic design's aesthetic explosion. The Web Design Museum shows this.

Nobody knew how to conduct browser-based graphic design. Web page design was undefined. No standards. No CMS (nearly), CSS, JS, video, animation.

Now is as good a time as any to challenge the internet’s visual conformity.

In 2018, everything is browser-based. Massive layouts to micro-typography, animation, and video. How do we use these great possibilities? Containerized containers. JavaScript-contaminated mobile-first pages. Visually uniform templates. Web design 23 years later would disappoint my younger self.

Our imagination, not technology, restricts web design. We're too conformist to aesthetics, economics, and expectations.

Crisis generates opportunity. Challenge online visual conformity now. I'm too old and bourgeois to develop a radical, experimental, and cutting-edge website. I can ask my students.

I taught web design at the Potsdam Interface Design Programme in 2017. Each team has to redesign a website. Create expressive, inventive visual experiences on the browser. Create with contemporary web technologies. Avoid usability, readability, and flexibility concerns. Act. Ignore Erwartungskonformität.

The class outcome pleased me. This overview page shows all results. Four diverse projects address the challenge.

1. ZKM by Frederic Haase and Jonas Köpfer

Frederic and Jonas began their experiments on the ZKM website. The ZKM is Germany's leading media art exhibition location, but its website remains conventional. It's useful but not avant-garde like the shows' art.

Frederic and Jonas designed the ZKM site's concept, aesthetic language, and technical configuration to reflect the museum's progressive approach. A generative design engine generates new layouts for each page load.

ZKM redesign.

2. Streem by Daria Thies, Bela Kurek, and Lucas Vogel

Street art magazine Streem. It promotes new artists and societal topics. Streem includes artwork, painting, photography, design, writing, and journalism. Daria, Bela, and Lucas used these influences to develop a conceptual metropolis. They designed four neighborhoods to reflect magazine sections for their prototype. For a legible city, they use powerful illustrative styles and spatial typography.

Streem makeover.

3. Medium by Amelie Kirchmeyer and Fabian Schultz

Amelie and Fabian structured. Instead of developing a form for a tale, they dissolved a web page into semantic, syntactical, and statistical aspects. HTML's flexibility was their goal. They broke Medium posts into experimental typographic space.

Medium revamp.

4. Hacker News by Fabian Dinklage and Florian Zia

Florian and Fabian made Hacker News interactive. The social networking site aggregates computer science and IT news. Its voting and debate features are extensive despite its simple style. Fabian and Florian transformed the structure into a typographic timeline and network area. News and comments sequence and connect the visuals. To read Hacker News, they connected their design to the API. Hacker News makeover.

Communication is not legibility, said Carson. Apply this to web design today. Modern websites must be legible, usable, responsive, and accessible. They shouldn't limit its visual palette. Visual and human-centered design are not stereotypes.

I want radical, generative, evocative, insightful, adequate, content-specific, and intelligent site design. I want to rediscover web design experimentation. More surprises please. I hope the web will appear different in 23 years.

Update: this essay has sparked a lively discussion! I wrote a brief response to the debate's most common points: Creativity vs. Usability