AI Art Weekly #39
Hello there, my fellow dreamers, and welcome to issue #39 of AI Art Weekly! 👋
Searching through AI Art Weekly has been a bit of a pain. So I’ve added a new Resources section that lists all past links and makes them searchable. I haven’t had the time to integrate links before issue #17 yet, but will do so over the weekend or next week. I’m also planning to add some categories to filter by tools, papers, tutorials, and so on. I hope you find it useful.
Let’s get to what’s new this week:
- Midjourney v5.2 got released
- Stable Diffusion XL 0.9 preview released
- MotionGPT can turn text to human motions and vice versa
- Diffusion with Forward Models reconstructs 3D scenes from a single image
- Blended-NeRF can manipulate existing NeRF scenes
- Interview with Nuclear Samurai and Jono K
- aiNodes Engine by deforum developer XmYx
- Texture Prompt for some blending magic
- And more
Cover Challenge 🎨
This weeks theme is sponsored by EXQUISITE WORKERS and the theme for next weeks cover is #AISurrealism. The reward is $100. Rulebook can be found here and images can be submitted here. Come join our Discord to talk challenges. I’m looking forward to your submissions 🙏
News & Papers
Midjourney v5.2 released
Midjourney v5.2 got released this week which introduces the following changes:
- Outpainting is finally here 🔥
- A new Aesthetic System with improved aesthetics and sharper Images, slightly improved coherence and text understanding
- A new “High Variation Mode” which makes all variation jobs much more varied
- And a new
/shorten
command which lets you ‘analyze’ a prompt and get suggestions on what words might not be doing anything and which ones might be key
I’ve compiled a thread with examples for all 4 new major changes on Twitter.
SDXL 0.9 preview released
The Stable Foundation released a new Stable Diffusion XL 0.9 preview on ClipDrop. According to their official announcement, we’re steps away from a public release of SDXL 1.0 for all!
Now SDXL might not have the same visual fidelity like Midjourney, but it’s still a very powerful tool, one I can’t wait to see what the open-source community will do with it once it gets released.
MotionGPT: Human Motion as Foreign Language
It’s been a while since we’ve some update in the field of Text-to-Motion. MotionGPT is the latest advancement, a user-friendly motion-language model to handle multiple motion-relevant tasks. By treating human motions as a language, the model is able to convert Text-to-Motion, Motion-to-Text and even predict future motions.
Diffusion with Forward Models: Solving Stochastic Inverse Problems Without Direct Supervision
Bringing movement into 2D images is something I’m super interested in. Diffusion with Forward Models is a able to reconstruct 3D scenes from a single input image. Additionally it’s also able to add small and short motions to images with people in them.
Blended-NeRF: Zero-Shot Object Generation and Blending in Existing Neural Radiance Fields
I haven’t shared NeRF-related progress in a while. Blended-NeRF is an interesting one. It lets you manipulate object textures as well as replace, blend and insert objects into an existing NeRF scene via text prompts or input images. Pretty cool.
More gems
- AvatarBooth: High-Quality and Customizable 3D Human Avatar Generation
- MagicBrush: A Manually Annotated Dataset for Instruction-Guided Image Editing
- DreamEdit: Subject-driven Image Editing
- MagicPony: Learning Articulated 3D Animals in the Wild
- AudioPaLM: A Large Language Model That Can Speak and Listen
@RenAI_NFT used the new MJv5.2 zoom out feature to create a zoom out animation by interpolating between the different zoom levels.
The opening title sequence for Marvel’s ‘SECRET INVASION’ show has been made with what looks like Midjourney v2, some img2img variations as well as deforum. Personally love the style but opinions are split on this one.
Interviews
Anna Dart and me got yet another two AI Surrealism interviews for you. This week with Nuclear Samurai and Jono K.
Tools & Tutorials
These are some of the most interesting resources I’ve come across this week.
Deforum developer XmYx released a new project called aiNodes Engine, a simple and easy-to-use Python-based AI image / motion picture generator node engine. I didn’t have the time to test it yet, but that’s definitely on my backlog.
@g_prompter has built a collection of prompt generators for Midjourney and LLMs. If you’re looking for inspiration, definitely worth a try.
@aisetmefree shared some texture / bg / wallpaper prompts which should be interesting for everybody embracing blending and img2img.
Meta released a HuggingFace demo of their MMS model that can transcribe and generate speech for 1000+ languages.
If you’re interested in prompt engineering, this guide by @dair_ai might be for you. Its focus is on prompting LLM’s and not diffusion models, but it’s an interesting read nonetheless.
And that my fellow dreamers, concludes yet another AI Art weekly issue. Please consider supporting this newsletter by:
- Sharing it 🙏❤️
- Following me on Twitter: @dreamingtulpa
- Buying me a coffee (I could seriously use it, putting these issues together takes me 8-12 hours every Friday 😅)
Reply to this email if you have any feedback or ideas for this newsletter.
Thanks for reading and talk to you next week!
– dreamingtulpa