HOME

A ongoing collection of thoughts, texts and experiments




29.01.2025
Farewell Dragon, hello Snake

I made some visual poems, inspired by the Chinese Lunar Year.




25.01.2025
Deepseek R1

The new chinese reasoning model Deepseek R1 is coming close to Open AIs o1-Model. And it’s open access under the MIT license.
You can test it on https://chat.deepseek.com or you can run it locally on Mac, Windows or Linux with the help of LM Studio.
Crazy times...

Deepseek R1 does well with task that require “reasoning” like math or coding. For creative writing Claude 3.5 Sonnet oder ChatGPT 4o might be a better choice.



25.01.2025
Design Research

Image AI Modeltest – One Prompt – Different AI models & Presets:
-> Midjourney
-> Flux (Pro-Ultra, Pro, Freepik Realism, Mystic) 
-> Stable Diffusion 3.5
-> Recraft V3
-> Ideogram 2.0
-> Frames (with different Presets)
-> Google Imagen
-> Adobe Firefly
-> Dall-E
-> Playground v3

Prompt:
Close-up portrait of a woman with long hair blowing in the wind, set against a beautiful landscape. The detailed skin and freckles are captured in soft light, creating an award-winning photograph. The image has been expertly color-graded, resulting in ultra-detailed photography in the style of Hasselblad medium format.

Results:
Midjourney good. Can be better with more Parameter Use.
The “male gaze” bias in Flux hard to ignore.
Recraft V3 was really a surprise.
Frames is kind of nice, but the women feel all the same, besides the ethical range (same noses, lips). Age must be really prompted in if wanted.
Midjourney V6.1 (with personal Profile)
Flux 1.1 Pro-Ultra
Flux 1.1 Pro
Flux Realism (Freepik)
Nvidia Sana
Mystic v2.5 Beta (Magnific, Flexible Preset) Flux-based
Mystic v2.5 Beta (Magnific, Editorial Portrait Preset) Flux-based
Stable Diffusion 3.5
SDXL Lightning 4step
Ideogram V2
Recraft V3 (no Preset)
Recraft V3 (Preset “realistic image)
Frames from Runway (no Preset)
Frames from Runway (Nordic Minimalism Preset)
Frames from Runway (High Contrast Warm Preset)
Frames from Runway (Terracotta Preset)
Frames from Runway (B&W Contrast Preset)
Luma Photon
Luma Photon (with Style Ref)
Google Imagen 3
Adobe Firefly
Dall-E
Playground v3






24.01.2025
Bro Technology?

Every Wednesday at 9pm I jump on Discord and tune in to Midjourney's “Office Hours”. I love this format. Midjourney CEO David Holz talks about the week's developments and upcoming features and then takes questions from the audience.

This week someone asked for his opinion on the new US administration. He said he thought they were “bro technology”. At least that was how it sounded over the Discourd audio. We were laughing about the pro/bro in the discord chat and a female user replied with the term “broligarchy”.

I've been thinking about this a lot over the last few days because I've noticed it on Linkedin and social media. "Bros come into the space and like each other and push each other for more clicks. I've seen “Top AI artists” lists made by Linkedin “bros” almost exclusively adding other "bros". "Bros" only commenting on "bros" posts and so on. 

So I think it's time to link to some non-bros (in no particular order)…
Tanya Rey
Samantha Borromeo
Jeanette Bergen
Stefanie Löwenstein
Gabriella “Tiny Rainboot”
Jumoke Fernandez
Nina Lemm
Nadia Piet
Julia Curty
Katja Dürig
Marianne Schmollgruber
Margrit Weber
Nici Jost
Nadine Friedel
Ramona Gschwend
Christina Tarquini
Soraya Jansen
Delia Guerriero
Holly Willis
Mieke Haase
Sabine von Bassewitz
Katrin Lange
Ellen Mathys
Adelina Arendarska
Sarah Kastrau
Claudia Bussjäger
Nina Habicht
Laura Pin
Nastassja Abel 

to be continued... 
(if I forgot you, message me)



22.01.2025
RunwayML “Frames”

I’ve had time to test Runways new image AI "Frames". The results vary greatly depending on the subject. It comes with a number of nice presets – and a very hefty price tag. Frames is only available with the Unlimited subscription ($95). I think it's nice to have if you already have the big subscription. Otherwise you can get similar results with Midjourney or Flux (via krea.ai, Freepik, Visual Electric or on Replicate or Fal) for less $$$. 

I also did video tests and often the model changed completely to another ethnicity or the nice blurry moods were gone in less than a second. So for now it's more of a nice-to-have but not a must have.







21.01.2025
Good Read

ZHdK Prof Felix Stalders article “Das Generative der Künstlichen Intelligenz” (german only).



20.01.2025
Living Encyclopedia: Large Nature Model

Refik Anadol Studio offered a 3-month free access for beta tester to their project ‘Living Encyclopedia,’ their ‘Large Nature Model,’ powered by Flux.

The website is fun to explore with visual and audio contents. It’s also possible to generate nature images.

-> https://lnm.dataland.art