Our great sponsors
-
-
InfluxDB
Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.
-
Context being that another AI art model (StableDiffusion) knows the names of many popular artists and can create images that sort of kind of look like their work. This terrified a lot of artists on Twitter who've now gone around harassing AI developers and claiming they're plagiarists, then simultaneously "umm this is all uncreative and ugly collages" and "this is going to take all our jobs".
Oddly, the main instigator turned out to be a "Pokemon in real life" fanartist who didn't notice he's already a professional plagiarist.
> So in that context, saying “horses stuck around when the automobile came” is true, but if you went up to a painter and said “hey, within your lifetime painting will see a 90% decline, stop being taught formally, disappear from daily life or awareness”.
The issue with this claim isn't automation replacing artists (though I don't think that will happen either due to Jevons' paradox) - it's just that AI generated images don't replace paintings because they aren't paintings! Print shops already exist and may have replaced you though.
> I’ve had a lot of struggles with this. I have a specific image in my head, I’m trying to prompt for it, and the AI just does not want to do it. The most trouble that I’ve had so far has been with trying to get a tavern running across the plain with chicken legs.
There's a general unfixable problem here, which is it's hard to be aligned with silly prompts without giving you silly output for "normal" prompts. That's also why they're complaining the model output has too safe composition - the developers are lucky they even got it to do that, it's better than random blobs of color and body parts like older models would generate.
But the picture they want probably is hiding somewhere in Midjourney's latent space; it's just a matter of finding a prompt that recreates it. One way to do that could be to sketch the picture you want some other way and run it through a reverse image-to-text notebook like https://github.com/pharmapsychotic/clip-interrogator.
Datasets can be manually curated to produce more aesthetic results if this becomes a real issue. For example, classifiers can predict whether an image is generated or not. You could adapt the process used to create laion-aesthetic[0] to remove generated images.
[0]: https://github.com/LAION-AI/laion-datasets/blob/main/laion-a...
DALL-E's docs for example mention it can output whole copyrighted logos and characters[1] and understands it's possible to generate human faces that are bear the likeness of those in the training data. We've also seen people recently critique Stable Diffusion's output for attempting to recreate artists' signatures that came from the commercial trained data.
That said by a certain point the kinks will be ironed out and likely skirt around such issues by only incorporating/manipulating just enough to be considered fair use and creative transformation.
[1] "The model can generate known entities including trademarked logos and copyrighted characters." https://github.com/openai/dalle-2-preview/blob/main/system-c...