AI Art Weekly #6
Welcome to issue #6 of AI Art Weekly. A newsletter by me (dreamingtulpa), to cover some of the latest happenings in the AI Art world.
Before we dive in, two updates:
- We’ve a new website: https://aiartweekly.com
- 2nd AI Art Weekly Cover Challenge announcement
2nd AI Art Challenge Cover 🎨
Winter is coming my fellow dreamers. And with the first season of House of Dragon wrapped up, I’m feeling inspired.
Two weeks ago, we held our first successful cover art challenge. This week I want to build upon that success and announce the 2nd AI Art Weekly Cover Challenge. But this time with a prize of $50 for the winner 👑.
To participate you must follow these rules:
- The theme is “ medieval dark fantasy ”, think Elden Ring, Berserk, A Song of Ice and Fire and so on.
- Max two submissions per participant.
- Submissions must be posted as a reply to the following tweet: https://twitter.com/dreamingtulpa/status/1585652935224168450
- Images must be vertical. The cover art has an A4 aspect ratio. So between 10:14 and 10:15 will result in the least amount of cropping.
- Submissions end next week on Thursday, 3rd of November 07:00 UTC.
- The winner allows me to mint and sell the image as an NFT. This helps to support the newsletter and future challenges like this.
After the deadline, I’ll pick my 4 favourites and create a poll to let the community decide on the final winner. The winner will receive the $50 prize and will be mentioned in the next issue. All other submitted images will be compiled into a gallery and put onto a website alongside their Twitter profiles.
I’m looking forward to all of your submissions 🙏
So with this out of the way, let’s dive into this weeks news.
If you want to support the newsletter, this weeks cover is available for collection on objkt as a limited edition of 10 for 2.50ꜩ a piece. Thank you for your support 🙏😘
Reflection: News & Gems
MidJourney V4 and Anime models
To start the week off, David Holz mentioned that MidJourney’s V4 model is just around the corner and they are apparently working on an Anime version as well. They’re currently running rating experiments to fine tune the style of their new model. Checkout their Discord announcements to help out. Speaking about Anime, NovelAI’s model is already creating stunning Anime output in case you’re into that.
In case you haven’t watched Stability AI’s Launch presentation, I got something exciting for you. They’re working on a Pro version of their DreamStudio suite which will enable creating animations and videos. We only got a glimpse of the user interface, but it looks like this will make this tech so much more accessible to a wider, less tech savvy, audience.
ARF: Artistic Radiance Fields
One of the most exciting papers I’ve come across this week is called ARF. ARF creates high-quality artistic 3D content by transferring the style of a single exemplar image, such as a painting or sketch, to NeRF and its variants. And the best thing, there is even code to play around with.
UrbandSound8K & Audio Style Transfer
Another exciting paper that got released this week is called Full-band General Audio Synthesis With Score-Based Diffusion (that’s a mouth-full). It’s an audio generator that is able to generate not only sounds, but also apply style transfers similar to what ARF does, but with audio files. Checkout their demo page for some insane examples like converting a singing woman into a barking dog.
Remember a few weeks ago when “all” the research labs were announcing Text-To-Video models but without any actual code or tools to play around with? Well this is about to change. GitHub user lucidrains implemented the Phenaki Video paper in pytorch. I haven’t seen an actual video generated with it so far, but we’re getting closer!
There is also another text2video implementation called… Text2Video. Although the image quality doesn’t compare to what diffusion models are putting out, what’s exciting about this one is that it’s fast. Let me quote:
As opposed to optimizing through an image generator model itself, which tends to be computationally heavy, the proposed approach computes the CLIP loss directly at the pixel level, achieving general content at a speed suitable for near real-time systems. The approach can generate videos in up to 720p resolution, variable frame-rates, and arbitrary aspect ratios at a rate of 1-2 frames per second.
This week I came across this gem by @SatsumaAudio and had to share it with you. He’s building a rhythm based platformer and uses AI generated art to create the assets.
It’s cool to see AI art in the real world. SuperChief hosted a gallery by AI artist z_kai in LA last friday. If you have a project that includes AI art that touches the physical world, reach out to me, I’d love to share more stuff like this in future issues.
My latest Deforum Stable Diffusion experiment. Basically blew an entire month worth of computing units into this and created 10 different versions. Key take away: apart from the audio reactive stuff, prompts are still the most important ingredient. So don’t slack on them!
Imagination: Interview & Inspiration
In this weeks interview of AI Art Weekly we talk to AmliArt. An artist I came across which feels like several months ago (but is only a few weeks in AI realm). She streams her AI adventures on Twitch and creates art that is filled with emotion from the entirety of the human spectrum. Let’s jump in.
[AI Art Weekly] Hey Amli, what’s your background and how did you get into AI art?
I have always been creative, but my background and training lay in the performative arts. I have been performing in the theater since I was eight years old and got my degree in acting. I pursued a career in the theater on the east coast and started a small theatrical company. However, I have always been someone who has wanted to do things my way, so when fate brought me to LA, and my acting trajectory turned toward film and television, I began to lose my passion for the art form. That, coupled with a disabling chronic illness, led me to give up the craft.
I do not talk about this much, but I mention it now because my upbringing in the world of storytelling profoundly affects how I create visual art. Whenever I come at a piece, I come from the deep desire to capture a story or moment of a story.
With the visual arts, I began self-training in painting and digital art in my early 20s as a personal pursuit. And as my joy and ability to act began to wane, I moved more and more toward visual storytelling. It was a way to continue to feed my creative outlet within my limitations.
My journey into AI is very similar. As my disease progressed, it became more difficult even to do the simple tasks of holding a paintbrush or stylist for long periods. However, discovering AI gave me my creativity back, allowing it to do some of the compositional heavy-lifting for me.
But I also discovered a partner in storytelling that I am sure I would have been drawn to even I did not have my physical limitations.
[AI Art Weekly] Do you have a specific project you’re currently working on? What is it?
I have a few projects I am working on, actually.
One is an animation series I call “Do You Love Me Now?” which explore the toxic narratives told by influencer and social media culture to the women in our society.
The second is a lighter series, capturing moments of love and sensuality portrayed in stories of fantastical characters.
[AI Art Weekly] What is your favourite prompt when creating art?
Bioluminescence. It feels silly when I say it, but it is in almost every single work I do. Even the white charcoal style works I have been working on lately. I like how it brings out a glow around objects and characters. Deep, I know.
[AI Art Weekly] What does your workflow look like?
I have a relatively slow workflow. I spend less time on prompt construction and more on manipulating the images I get from the AI.
I usually start with what I call an AI sketch. This means I will use a basic prompt about the general idea I am going for (usually with
sketch in there) and get rough images. Then, I will find one that inspires me, take it out, and work with it in photoshop to add or remove aspects I want.
Then I will take that image and rerun it through the AI, changing the prompt based on what the image is morphing into. Then I wash and repeat until I feel the story is complete. I have a few process videos on this you can check out.
With animation, I am also hands-on. First, I create an initial image I want the animation to run off, where I want the story to start. Then I will spend time crafting prompts for each beat of the story. But rather than walking away, I will stay with the animation as it runs, stopping the process if it starts to go in a direction I don’t want. Finally, I will tweak the prompt to help guide it where I want and restart the process. This gives me more control over the story I want to tell.
[AI Art Weekly] How do you feel AI (art) will be impacting society?
Wow, that is a whopper of a question that I’m not sure can be answered at any given time. It is an ever-evolving question, and evolving at such a rate as to be mindblowing. I co-host a space with @bl_artcult called “The New Renaissance” where we address this and other questions every Wednesday at 12 pm PT.
My HOPE with AI art is that it will completely change our cultural, and artistic narrative by helping break down the gatekeeping that has surrounded the arts since… well… ever.
I think we often forget that the pursuit of art is a pursuit of the privileged. You need money and time. Even if you are self-teaching, you need time, which the disenfranchised communities of our society do not have.
When Stability brought AI to the masses, I remember a short debate over the excess of naked white women portrayed by the tool. People called the tool racist or sexist (as if it could be). What many people don’t think about is how these tools mirror what we as a society have believed art “looks like.”
In a nutshell, they are trained on our artistic history and therefore overtrained on naked white women since privileged white men have created most art. Of course, these are all broad statements, and there is a lot to unpack there, which we could talk about for hours, but I won’t for the sake of the readers time.
My point is that when these tools first came out, I was hopeful that they would break open the barrier-of-entry to art and that we would finally see the untold stories and perspectives of an entire swath of our society that has so far been missing.
I still believe that these tools could help in that direction, at the very least I see them being of great value to the disabled. But the general trend now is one of once again embracing monetary gatekeeping. As a result, the tools and equipment needed to run AI are becoming less and less affordable. But, I do believe the barrier is still lower than it was, so I am still hopeful.
[AI Art Weekly] Who is your favourite artist?
I don’t know if I have a favourite artist, as much as I have favourite pieces. Currently, my favourite piece is the following by @_MemoryMod_.
[AI Art Weekly] Anything else you would like to share?
Ohh don’t give me an open-ended question like this; I will ramble!
I will end with two things that I believe are causing burnout in the community.
There is a lot of vitriol from both sides of the AI art debate and we need to start stepping back from interacting with these discussions. Instead of trying to defend ourselves and shout our ‘worth’ to the masses, we should show our worth through our work. Every time we interact with those spouting hate on social media, all we are doing is fueling their fire and upsetting ourselves. Obviously, some conversations still need to happen, and if there are people genuinely having those and not just spouting hate, then I do believe engagement is essential. The ethical use of these tools will be an ongoing and vital discussion, for example. But when it comes to trolls and hate - block, block, block. It is simply not worth it.
The pace of this technology can be overwhelming. It can feel like you will never catch up, or you are stupid for being unable to figure it out. But we are ALL struggling through every new update, even those of us you may look up to and think, ‘wow, that person has it all together and is doing such great work, and I could never be like them.’ Trust me, you are like them.
Clumsily curious, viciously vulnerable. Digital artist learning and educating about image synthesis tools.
I will do a shameless plug here. I host a twitch space on Tuesdays at 6pm PT called “Let’s Figure it out” where it’s just me being a hot mess, banging my head against the wall, trying to figure out new AI tools. It is important to break down narratives of effortless perfection, to let people know that we are all struggling, and to use our community as our backbone. We should share, learn, and educate together where we can. This is an art form built by community after all, and it is one of the most beautiful things about it.
[AI Art Weekly] Bonus question: Can you tell me about #baconwave ?
Hahaha ya. The provenance of baconwave 😂. It’s origin story might not be the epic tale people expect with such a prolific cultural revolution.
I made the following comment in a tweet I had made asking about deforum MATHS:
Then @coffeevectors said this in response:
And @weirdmomma created the first image:
And then coffee, momma and kyrickyoung just started spiraling off each other. I believe @KyrickYoung then posted the first image with the hashtag the next day. And viola, a movement was born.
Each week we share a style that produces some cool results when used in your prompts. This weeks featured style is
in the style of pascale campion and olivier bonhomme.
Creation: Tools & Tutorials
These are some of the most interesting resources I’ve come across this week.
This is the fine-tuned Stable Diffusion model trained on images from the TV Show Arcane. Use the tokens
arcane style in your prompts for the effect.
I came across Arcane Diffusion from a tweet by @rainisto where he explains how he uses Dreambooth to train SD new styles. In case you want to go a bit deeper, Suraj Patil created an analysis on their experiments with DB that might be interesting for you.
MotionDiffuse is a Text-To-Motion model which lets you generate human animation sequences. There is no export option to your favourite 3D tool yet, but they are working on it
I’ve put together a simple web utility that let]s you easily put together and adjust prompt keyframes for animations similar like this. Not rocket science, but helpful either way.
Last week I shared a Google Colab tutorial which let’s you generate audio tracks with Mubert’s Text-To-Audio model. This week they added a Hugging Face space, which makes generating audio tracks even easier now.
And that my fellow dreamers, concludes this weeks AI Art weekly issue. Please consider supporting this newsletter by subscribing and sharing. Let me know on Twitter if you have any feedback and ideas.
Thanks for reading and see you next week!