[13:37 Thu,21.September 2023 by blip] |
OpenAI has introduced the latest version of its diffusion-based AI image generator DALL-E - it is now closely integrated with ChatGPT, which brings several advantages. For example, prompt specifications should be followed much more accurately than before, and the display of text in the generated images should also succeed better. ;) Images generated by DALL-E 3 In fact, DALL-E 3 is said to be "built natively on ChatGPT", although there are unfortunately no exact technical details about the model architecture or the training. The somehow multimodal approach creates a different relationship between speech and image, resulting among other things in more influence on image generation.
DALL-E 3 will even run directly in the interface of ChatGPT, so that the text generator can immediately formulate the exact prompts. All you have to do is ask for an image and ChatGPT will act as an intermediary to DALL-E 3, making cumbersome prompt engineering a thing of the past, according to OpenAI. Who would like, can enter naturally also even a detailed description of its picture idea. The sample images selected by OpenAI - currently DALL-E 3 is still in closed beta - indeed show a great closeness between prompt and generated image: ;) The following image, in turn, is based on this prompt - note, by the way, the impeccable rendering of the hands: "A middle-aged woman of Asian descent, her dark hair streaked with silver, appears fractured and splintered, intricately embedded within a sea of broken porcelain. The porcelain glistens with splatter paint patterns in a harmonious blend of glossy and matte blues, greens, oranges, and reds, capturing her dance in a surreal juxtaposition of movement and stillness. Her skin tone, a light hue like the porcelain, adds an almost mystical quality to her form." ;) Text is also inserted correctly, mainly when given verbatim in the prompt; this did not work before. ;) Basically, gibberish can still be found in images of DALL-E 3. Thus the prompt for the following poster contained the default "The bottom text reads, Explore Venus: Beauty Behind the Mist" - this was taken over largely, but not completely, while the small print was freely fabricated as before. ;) Laut OpenAI sind mehrere Filter im neuen KI-Bildgenerator implementiert, nicht nur um die Darstellung von Gewalt u.ä. zu verhindern. Es soll demnach ebenso wenig möglich sein, Bilder von bekannten Persönlichkeiten zu erstellen, zumindest indem ihr Name im Prompt genannt wird. Auch sollen sich keine Bilder mehr im Stile von noch lebenden Künstlern generieren lassen. Darüberhinaus soll das Unternehmen an einer internen Kennung arbeiten, um künftig erkennen zu können, welche Bilder mit DALL-E 3 generiert wurden. DALL-E 3 soll Anfang Oktober für ChatGPT Plus und Enterprise Kunden zugänglich werden (also kostenpflichtig).  more infos at bei openai.comdeutsche Version dieser Seite: DALL-E 3 ist da und generiert exaktere KI-Bilder inkl. Text dank ChatGPT-Integration |
 |
|
|