I have read that simple img2img models can effectively remove glaze and nightshade. Still hasn't been addressed by the paper's authors. :/
October 30, 2026 at 6:04 AM
I have read that simple img2img models can effectively remove glaze and nightshade. Still hasn't been addressed by the paper's authors. :/
Except these tools won't generally substantially reproduce an existing work. As I've previously explained, this often will only occur when someone specifically uses an existing work as a basis for their AI art through img2img.
November 1, 2026 at 4:30 AM
Except these tools won't generally substantially reproduce an existing work. As I've previously explained, this often will only occur when someone specifically uses an existing work as a basis for their AI art through img2img.
It's hard to answer that question since my workflow is kinda chaotic and inconsistent. There is no simple prompt that will give you the same result. For this one I was using tools like img2img and controlnet + manual editing
October 31, 2026 at 3:25 PM
It's hard to answer that question since my workflow is kinda chaotic and inconsistent. There is no simple prompt that will give you the same result. For this one I was using tools like img2img and controlnet + manual editing
This tracing analogy is so absurd. When one traces, they're directly reproducing a work verbatim. Producing AI art would only be remotely comparable to tracing if you were to use img2img to make changes to existing works, and that's more like editing than tracing.
October 31, 2026 at 3:29 PM
This tracing analogy is so absurd. When one traces, they're directly reproducing a work verbatim. Producing AI art would only be remotely comparable to tracing if you were to use img2img to make changes to existing works, and that's more like editing than tracing.
I've quite literally explained why. A diffusion model doesn't contain imagery. So, how could the inference process trace an existing work it doesn't have access to? The only time it's comparable to tracing is if a particular user provides an image as a basis for their work via img2img.
October 31, 2026 at 3:23 PM
I've quite literally explained why. A diffusion model doesn't contain imagery. So, how could the inference process trace an existing work it doesn't have access to? The only time it's comparable to tracing is if a particular user provides an image as a basis for their work via img2img.
img2img from three others including real things and other digital art. I do minimal generic prompts to let the inputs combine as weirdly as possible :-) (takes a lot of curation. i feel like i did with my first digital cameras "oh wow infinite film and no running to the drugstore!")
October 31, 2026 at 2:55 AM
img2img from three others including real things and other digital art. I do minimal generic prompts to let the inputs combine as weirdly as possible :-) (takes a lot of curation. i feel like i did with my first digital cameras "oh wow infinite film and no running to the drugstore!")
By img2img model it sounds kinda like another AI in itself but what I mean is even more simpler, like, what if before training they... compress the image again? Like how JPEG does it? High level details aren't lost, but the low level details which I think is where the glaze resides at, are gone lol
October 30, 2026 at 10:53 AM
By img2img model it sounds kinda like another AI in itself but what I mean is even more simpler, like, what if before training they... compress the image again? Like how JPEG does it? High level details aren't lost, but the low level details which I think is where the glaze resides at, are gone lol
I might have had some respect if Art had learnt how to do the prompting himself and used Img2Img on his own sketches and rough work. This is just being a total lazy arse.
October 25, 2026 at 8:07 PM
I might have had some respect if Art had learnt how to do the prompting himself and used Img2Img on his own sketches and rough work. This is just being a total lazy arse.
Looks like free Gemini and paid Gemini is slowly making the gap between images from the pro model and the 2.1 model even further apart. So many sneaky ways they force the image gen model doing only kinda correct img2img.
October 24, 2026 at 8:05 PM
Looks like free Gemini and paid Gemini is slowly making the gap between images from the pro model and the 2.1 model even further apart. So many sneaky ways they force the image gen model doing only kinda correct img2img.
the scale ambiguity is the fun here, reads like aerial coastline and microscope slide at once. img2img from a texutre photo, or pure prompt?
October 22, 2026 at 7:23 PM
the scale ambiguity is the fun here, reads like aerial coastline and microscope slide at once. img2img from a texutre photo, or pure prompt?
This shouldn't even be legal.
And everyone has a camera in their pocket, so you go img2img, not prompt to image, as a bare minimum.
But that shouldn't be legal.
www.bbc.co.uk/news/article...
And everyone has a camera in their pocket, so you go img2img, not prompt to image, as a bare minimum.
But that shouldn't be legal.
www.bbc.co.uk/news/article...
Just Eat apologises for AI food mistakes after social media ridicule
A spokesperson said it was removing inaccurate AI images and pausing the rollout of the tool.
www.bbc.co.uk
October 22, 2026 at 10:42 PM
This shouldn't even be legal.
And everyone has a camera in their pocket, so you go img2img, not prompt to image, as a bare minimum.
But that shouldn't be legal.
www.bbc.co.uk/news/article...
And everyone has a camera in their pocket, so you go img2img, not prompt to image, as a bare minimum.
But that shouldn't be legal.
www.bbc.co.uk/news/article...
qwen q2 img2img is making like, cracked oil paint looking images and it's kind of cool
October 21, 2026 at 5:28 PM
qwen q2 img2img is making like, cracked oil paint looking images and it's kind of cool
No, they were just wrong. The inference software doesn't actually use imagery in order to produce new works, unless specifically supplied by the user via something like img2img. A diffusion model isn't imagery.
October 21, 2026 at 12:41 AM
No, they were just wrong. The inference software doesn't actually use imagery in order to produce new works, unless specifically supplied by the user via something like img2img. A diffusion model isn't imagery.
Models can memorize how to reproduce specific works, but this is quite rare. When you see a model substantially reproduce an exist work, it's often because the artist used an existing work as a basis for their AI art. This is done using features like img2img.
October 21, 2026 at 8:17 AM
Models can memorize how to reproduce specific works, but this is quite rare. When you see a model substantially reproduce an exist work, it's often because the artist used an existing work as a basis for their AI art. This is done using features like img2img.
Interesting to see how Gemini had the same prompt but word soup on 1 results in a more "lore" leaning Shimmer, while result 2 returned a more img2img result. Both are pretty cute, as I wasn't sure which one to post.
October 12, 2026 at 3:24 AM
Interesting to see how Gemini had the same prompt but word soup on 1 results in a more "lore" leaning Shimmer, while result 2 returned a more img2img result. Both are pretty cute, as I wasn't sure which one to post.
Sometimes when i’m bored i’ll chop up a piece of album art into pieces and then have a basic img2img model generate an appropriate image based on nothing but the titles of each song, and boy is it ever an uphill battle to stop it from trying to add text if there’s any text in the original
October 10, 2026 at 7:52 PM
Sometimes when i’m bored i’ll chop up a piece of album art into pieces and then have a basic img2img model generate an appropriate image based on nothing but the titles of each song, and boy is it ever an uphill battle to stop it from trying to add text if there’s any text in the original
in 2022 i was reluctant to call myself an ai artist, i felt like more of a director and curator
even as i moved to spending hours on img2img, photobashing, even spending weeks on comics arrangement and collaging
and then the discourse got bad so
im an ai artist fuck it lol
(comic from 2023)
even as i moved to spending hours on img2img, photobashing, even spending weeks on comics arrangement and collaging
and then the discourse got bad so
im an ai artist fuck it lol
(comic from 2023)
October 10, 2026 at 1:33 PM
in 2022 i was reluctant to call myself an ai artist, i felt like more of a director and curator
even as i moved to spending hours on img2img, photobashing, even spending weeks on comics arrangement and collaging
and then the discourse got bad so
im an ai artist fuck it lol
(comic from 2023)
even as i moved to spending hours on img2img, photobashing, even spending weeks on comics arrangement and collaging
and then the discourse got bad so
im an ai artist fuck it lol
(comic from 2023)
yeah, that's where I'm not sure the model I want to talk to really exists these days. It's all just multi-modal heads grafted onto LLMs. Plus they're all good at writing text in images now (well, kinda, they love it too much now). I think I want a good old pure text2img model. maybe even img2img
October 8, 2026 at 4:36 AM
yeah, that's where I'm not sure the model I want to talk to really exists these days. It's all just multi-modal heads grafted onto LLMs. Plus they're all good at writing text in images now (well, kinda, they love it too much now). I think I want a good old pure text2img model. maybe even img2img
Ja. Flux2 kann andere Sachen. Deshalb ist es sinnvoll die Modelle zu kombinieren.
Krea 2 Txt2Img
Flux2klein9b für image to image mit Styletransfer
Qwen 2.1 für Img2img, inpaint und outpaint.
Krea 2 Txt2Img
Flux2klein9b für image to image mit Styletransfer
Qwen 2.1 für Img2img, inpaint und outpaint.
October 5, 2026 at 2:34 PM
Ja. Flux2 kann andere Sachen. Deshalb ist es sinnvoll die Modelle zu kombinieren.
Krea 2 Txt2Img
Flux2klein9b für image to image mit Styletransfer
Qwen 2.1 für Img2img, inpaint und outpaint.
Krea 2 Txt2Img
Flux2klein9b für image to image mit Styletransfer
Qwen 2.1 für Img2img, inpaint und outpaint.
Garuda Nix, Img2Img AI, WanderWallet, Quetta y Plasma One plantean otra forma de trabajar, crear y pagar mientras viajas. Cinco innovaciones, una misma pregunta: ¿cuánto control estamos dispuestos a asumir?.
https://smpl.is/an4dn
https://smpl.is/an4dn
October 3, 2026 at 2:00 PM
Garuda Nix, Img2Img AI, WanderWallet, Quetta y Plasma One plantean otra forma de trabajar, crear y pagar mientras viajas. Cinco innovaciones, una misma pregunta: ¿cuánto control estamos dispuestos a asumir?.
https://smpl.is/an4dn
https://smpl.is/an4dn
Gemini's img2img flow has actually become my "shitpost photoshop" tool of choice once it got good at kitbashing several images, but that is not a high value use case for me.
October 1, 2026 at 11:42 AM
Gemini's img2img flow has actually become my "shitpost photoshop" tool of choice once it got good at kitbashing several images, but that is not a high value use case for me.
Img2Img AI reúne edición con texto, restauración de fotos, eliminación de objetos, cambios de fondo y conversión de imágenes a 3D. El editor incluye historial de versiones.
Descubre cómo funciona:
https://smpl.is/an4dd
Descubre cómo funciona:
https://smpl.is/an4dd
September 30, 2026 at 2:00 PM
Img2Img AI reúne edición con texto, restauración de fotos, eliminación de objetos, cambios de fondo y conversión de imágenes a 3D. El editor incluye historial de versiones.
Descubre cómo funciona:
https://smpl.is/an4dd
Descubre cómo funciona:
https://smpl.is/an4dd