|
About:
|
Stable ai diffusion image generated by ai porn generators - https://pornpen2.com/ - faces. Image courtesy of stability ai. Stability ai recently announced the public launch of its text-to-image-based stability diffusion ai image generator that can run on consumer gpus to generate a custom image in seconds . The internet is kind of freaking out. People think: should it be legal to create art derived from copyrighted works, and will artists lose their jobs if ai can draw its own image in 2 minutes? In addition, ai needs bias to avoid being baked by getting mostly images of white people from the internet, as well as many other bias and image representation issues that lead to police profiling, biased facial recognition systems, and ai generators. The ability to be used to create fake revenge porn images, among other malicious content. A stable diffusion image of an ai-generated image of a war zone or natural disaster. Image courtesy of stability ai. This image generator looks to be quite efficient and impressive based on the published images we have posted here for you, and we are sure that companies already leaning towards stock images will can shift in this direction for custom artwork for free. However, don't grab the pearls just yet, because we have news for you. Ai is notoriously bad at drawing certain things. For example, hands. Yes, this ai image generator can create custom images that rival evangelion at times. What ai can't do very well yet is draw houses in the background of your requested image, or spacecraft control details: they come out pixelated and warped, with a roofline halfway back from the walls, or buttons twisted into strange shapes. (See penultimate image below for an example of a really great image that still doesn't make sense on closer inspection). Getting the ai to understand certain relationships is quite painful. Asking an ai generator like dalle to draw a picture of a connective tissue disease? My eyes. We asked dalle about "ehlers-danlos syndrome", an inherited connective tissue disorder that results in fragile collagen, based on an internet user who posted what appeared to be art created from a request that resembled a human ribcage and nervous system writhing like a spiral staircase. Interesting. Dalle created "ehlers-danlos syndrome", an inherited collagen disorder that the ai generator misinterpreted as affecting the shape of the arms and legs. Or something like that. Dalle drew distorted images of deformed hands. Or was it the legs? Some images show bruising, others show visible veins, which is likely an interpretation of some of the symptoms of ehlers-danlos disease, which may include light bruising or visible veins. However, it was hard to tell what the images were generating due to the distortion. Could it be a rash or a hematoma? In any case, ai can only associate relationships between images with detailed hints and good input, so a lot depends on what you put into the generator, what you get. The new stable diffusion generator can use more detailed clues and should have better relationship associations, and was trained on a model based on the work of compvis and runway in their popular latent diffusion model with additional knowledge from conditional diffusion models by stable diffusion generative ai lead developer katherine crowson, dall-e 2 from open ai, imagen by google brain. So much more than just dalle. Dalle recreates william carlos williams' famous poem: "so much depends on a red car..." We fed to the image generator, an excerpt from a poem by william carlos williams about a red wheelbarrow full of chickens in dalla. He put a raw chicken patty in a wheelbarrow, bringing new layers of meaning to english learning everywhere. We think most of the artists will keep their main jobs for now. We tried again: a chicken melting in a wheelbarrow? A chicken with wheelbarrow legs. Is stable diffusion much different? It looks like it is, but it's not yet clear how detailed the hints are or how much filtering went into the posted images.Developers are given first access to test the generator, so let us know. What seems to be a significant difference is that the core dataset was trained on laion-aesthetics, a subset of laion 5b that will be released soon. Laion-aesthetics was created with a new clip-based model that filtered laion-5b based on how "beautiful" the image was, based on ratings from stable diffusion alpha testers. So there is a human level of interpretation and feedback about image fit that can teach this model to be a bit more... Tasteful. But this is entirely up to the testers, which raises additional potential legal and ethical questions for the future use of ai image generation. Stable, diffuse image of ai-generated fantasy art. Image courtesy of stability ai. Release details of the stable diffusion ai image generator The developers of this ai generator say: “because these models were trained on image-text pairs from an extensive web page, the model may reproduce some social biases and create insecure content, so open mitigation strategies, as well as open discussion of these biases, can bring everyone into this conversation. Learn more about the strengths and limitations of the model in the model card." This really raises the issue of perpetuating the current issues that exist in the dataset available to feed to image generators. In fact, all sorts of things can come up. Problems other than police profile, facial recognition technology bias, or underrepresentation. Dermatologists have recently faced retribution for failing to show images of darker skin with rashes and other skin complaints. That it is easier to see rashes and scarring on lighter skin, which is why medical textbooks used photographs of only fair skin.But skin complaints appear differently on darker skin, for example, the rash appears gray on dark skin and red on fair skin, so colored patients suffer from a lack of representation in medical texts, which can lead to misdiagnosis or undertreatment of medical complaints. This is one of the many potential hurdles ai must overcome in order to create meaningful and ethical content that can be used in more than just fantasy art or stock images. However, the new image generator also seems to have overcome many of the previous challenges to produce great images from certain cues. Stable ai-generated diffuse face image. Image courtesy of stability ai. You should see this for yourself. The new image generator has been tested by over 10,000 beta testers creating 1.7 million images per day and includes a feature that allows you to specify what you don't want in the output, which is still a work in progress and may have great value in producing the desired results (if you understand what you are getting into at all: we have grown the image of a chicken melting in a wheelbarrow).
|