Hannu Töyrylä
@htoyryla.bsky.social
An engineer turned humanities scholar turned visual artist. AI art in my own way since 2015. Or rather... my own art using photography, AI and image processing. Mostly code my own tools myself.
Kriitikon arvio on toki hänen omansa, eikä sen kanssa väitellä, mutta jos kriitikko esittää virheellistä tietoa taitelijan työskentelystä, siihen voi toki vastata.
vimeo.com/1126585537/4...
vimeo.com/1126585537/4...
Kuinka taiteeni oikeasti syntyy?
Olen vuodesta 2015 tehnyt taidetta kehittelemällä omaa kuvankäsittelyä johon yhdistelen kuvallista tekoälyä. Nyt kun isot kielimallit…
vimeo.com
October 12, 2025 at 1:52 PM
Kriitikon arvio on toki hänen omansa, eikä sen kanssa väitellä, mutta jos kriitikko esittää virheellistä tietoa taitelijan työskentelystä, siihen voi toki vastata.
vimeo.com/1126585537/4...
vimeo.com/1126585537/4...
Saaran ja minun yhteinen näyttely Rovaniemellä. Siinä sekoittuvat Saaran värikenttämaalaus ja oma tapani käsitellä kaupunkimaisemia, jossa tällä kertaa sekoittuvat niin Rovaniemen keskustan näkymät, Saaran maalaukset kuin myös itse galleriatila.
Avajaiset 26.9. 2025 klo 18-20
Avajaiset 26.9. 2025 klo 18-20
September 17, 2025 at 11:31 AM
Saaran ja minun yhteinen näyttely Rovaniemellä. Siinä sekoittuvat Saaran värikenttämaalaus ja oma tapani käsitellä kaupunkimaisemia, jossa tällä kertaa sekoittuvat niin Rovaniemen keskustan näkymät, Saaran maalaukset kuin myös itse galleriatila.
Avajaiset 26.9. 2025 klo 18-20
Avajaiset 26.9. 2025 klo 18-20
Weird stares.
(at ArtSibbe25 exhibition, Sipoo, Finland, today)
(at ArtSibbe25 exhibition, Sipoo, Finland, today)
August 10, 2025 at 7:05 PM
Weird stares.
(at ArtSibbe25 exhibition, Sipoo, Finland, today)
(at ArtSibbe25 exhibition, Sipoo, Finland, today)
One piece from my Complexions series now on auction as NFT at gallery.manifold.xyz/0x1a722f5890...
May 2, 2025 at 2:31 PM
One piece from my Complexions series now on auction as NFT at gallery.manifold.xyz/0x1a722f5890...
Modified my urdiffusion to use EULER scheduler.
Urdiffusion is for experimenting with training small models on limited private datasets overnight. Focused rather than generic aesthetics.
Urdiffusion is for experimenting with training small models on limited private datasets overnight. Focused rather than generic aesthetics.
April 24, 2025 at 7:58 AM
Modified my urdiffusion to use EULER scheduler.
Urdiffusion is for experimenting with training small models on limited private datasets overnight. Focused rather than generic aesthetics.
Urdiffusion is for experimenting with training small models on limited private datasets overnight. Focused rather than generic aesthetics.
What if you train a diffusion model from scratch using only eight images in the training set, and train it only for half an hour.
April 12, 2025 at 8:10 AM
What if you train a diffusion model from scratch using only eight images in the training set, and train it only for half an hour.
April 3, 2025 at 6:22 PM
Work in progress viewable live at liipetti.net/htoyryla/z12...
Hannu Töyrylä 2025 WIP
liipetti.net
March 13, 2025 at 3:02 PM
Work in progress viewable live at liipetti.net/htoyryla/z12...
Points of view.
A set of selfies run through several self developed processes. Soundtrack my own music.
A set of selfies run through several self developed processes. Soundtrack my own music.
March 11, 2025 at 4:56 PM
Points of view.
A set of selfies run through several self developed processes. Soundtrack my own music.
A set of selfies run through several self developed processes. Soundtrack my own music.
My 10 years in AI art have left a trace in my Twitter/X account. This Grok summary captures the essence if it, even if the details like GPU cable can be excused as poetic license.
AI art today is something quite different, I am still going my own ways.
AI art today is something quite different, I am still going my own ways.
March 11, 2025 at 3:10 PM
My 10 years in AI art have left a trace in my Twitter/X account. This Grok summary captures the essence if it, even if the details like GPU cable can be excused as poetic license.
AI art today is something quite different, I am still going my own ways.
AI art today is something quite different, I am still going my own ways.
Nyt menee mielenkiintoiseksi. Viime vuonna taidemaalarivaimoni Saara otti maalauksiensa osaksi tekoälyllä tekemiäni (ja enimmäkseen syrjään laittamiani kuvia). Nyt itse otan alunperin maisemakollaaseja kehittelevään prosessiini mukaan Saaran maalauksia, juuri niitä joissa on kuviani mukana.
February 4, 2025 at 12:29 PM
Nyt menee mielenkiintoiseksi. Viime vuonna taidemaalarivaimoni Saara otti maalauksiensa osaksi tekoälyllä tekemiäni (ja enimmäkseen syrjään laittamiani kuvia). Nyt itse otan alunperin maisemakollaaseja kehittelevään prosessiini mukaan Saaran maalauksia, juuri niitä joissa on kuviani mukana.
A fragment. Around here. With soundtrack.
January 26, 2025 at 4:13 PM
A fragment. Around here. With soundtrack.
How is it possible that everything I am deeply interested in the long run turns out to be controversial.
Perhaps everything is.
Perhaps everything is.
January 1, 2025 at 8:19 PM
How is it possible that everything I am deeply interested in the long run turns out to be controversial.
Perhaps everything is.
Perhaps everything is.
Some people claim that a generator must be capable of all possible visual styles to allow the artist to find his own style. I prefer to work with generators with a limited range of expression. They are difficult to work with but when you succeed you know it was worth while.
December 7, 2024 at 2:01 PM
Some people claim that a generator must be capable of all possible visual styles to allow the artist to find his own style. I prefer to work with generators with a limited range of expression. They are difficult to work with but when you succeed you know it was worth while.
Working on my own GAN again... this is one way to ensure that there's nobody's images involved in the process but mine.
December 4, 2024 at 6:06 PM
Working on my own GAN again... this is one way to ensure that there's nobody's images involved in the process but mine.
Poorly trained (or designed?) GAN used a generative tool, continued.
December 1, 2024 at 6:44 PM
Poorly trained (or designed?) GAN used a generative tool, continued.
A GAN that fails to learn correct features from the training set as a generative process on its own right.
November 28, 2024 at 8:04 PM
A GAN that fails to learn correct features from the training set as a generative process on its own right.
First train a GAN on something... say faces. Then you start a training from this model but using a different dataset. Monitor how the output changes from faces into something else, and save frequently.
November 26, 2024 at 7:30 PM
First train a GAN on something... say faces. Then you start a training from this model but using a different dataset. Monitor how the output changes from faces into something else, and save frequently.
You have 15 new followers.
aka Return of the son of the NQGAN. Working on to revive my favorite tools from the GAN era.
aka Return of the son of the NQGAN. Working on to revive my favorite tools from the GAN era.
November 26, 2024 at 2:12 PM
You have 15 new followers.
aka Return of the son of the NQGAN. Working on to revive my favorite tools from the GAN era.
aka Return of the son of the NQGAN. Working on to revive my favorite tools from the GAN era.
Yesterday evening I felt like coding a new GAN implementation using ResNets. Checked ChatGPT for basic code for G and D. There were some errors, but it was a good exercise to find them. After figuring out good learning rates it worked quite well. Now I will start adding some additional tricks...
November 21, 2024 at 11:24 AM
Yesterday evening I felt like coding a new GAN implementation using ResNets. Checked ChatGPT for basic code for G and D. There were some errors, but it was a good exercise to find them. After figuring out good learning rates it worked quite well. Now I will start adding some additional tricks...