OpenAI may, in whispers, be crafting the successor to its illustrious DALL-E AI text-to-image generator, whispers of alpha tests echoing through the digital ether, as revealed by The Decoder.
A shadowy figure on the Discord shores divulged his clandestine rendezvous with the forthcoming OpenAI marvel, christened DALL-E 3. Emerging like a specter in May, he regaled the attentive Discord congregation with tales of his sojourn as an alpha tester for OpenAI, delving into the unexplored realms of a new AI image model. The images he conjured at that time were like spells woven into pixels.
Behold, a revelation like no other – Image Generation of Unprecedented Brilliance! The May iteration of the alpha test possessed the gift to birth images of myriad aspect ratios within the creative vault of the model. Enter the realm of MattVidPro AI, the bard of YouTube, showcasing creations forged in the crucible of a 16:9 aspect ratio. This version showcased an unparalleled mastery in textual alchemy, a domain where even the titans of the field – Stable Diffusion and Midjourney – falter.
Witness, in awe, the visages born from the dreamstuff of DALL-E: text intertwined with the sinew of brick walls, neon signs heralding cryptic words, billboards winking in cityscape radiance, cakes bedecked in edible poetry, and names inscribed upon mountainous hearts. For the domain of people, DALL-E reigns supreme – behold a woman feasting upon spaghetti at a festive gathering, as seen through the unblinking eye of a fish.
Returning to the hallowed halls of Discord in the midst of July, the enigmatic leaker unfurled new images and secrets. Enswathed in the cloak of a “closed alpha”, with a retinue of some four hundred subjects, he claimed entry through an enigmatic invitation in the inbox. A veteran of trials past, of the original DALL-E and its sequel, the shadows whisper that this dark journey may lead us to DALL-E 3, though the stars remain silent.
The passage of time has wrought a transformation within the depths of the model. The leaker reveals this metamorphosis by unveiling images born from the same spark, showcasing the burgeoning might of DALL-E 3. The prompt unfurls a tapestry of wonder: a painting of a pink jester bestowing a high-five upon a somber panda in a cycling ballet amidst a fogous forest. The wheels – nay, the bicycles themselves – fashioned from the curdled embrace of cheese, and the earth beneath muddied by unseen hands. The panda, a creature of ire.
In the May iteration, the scene unfolds in shades of truth, though the hands falter in their connection, and the wheels turn yellow rather than golden with cheese. Yet, in the July iteration, a symphony of detail envelops the jester and the panda in a timeless dance of camaraderie, the wheels fashioned from the very essence of indulgence.
Yet further afield, in the lands of Midjourney, the jester fades into obscurity, the pandas forsake their cycles for metal steeds, roads replacing the mire, joy supplanting fury.
Grand images of the DALL-E 3 July alpha parade before our eyes, a breathtaking vista of future dreams. Yet, the unshackled alpha harbors shadows of a darker hue, whispers of scenes best left untoched – of violence, nudity, and logos prized by corporate titans.
Gaze upon the tapestry unfurled: a cruel anime specter, a visage of Westeros clad in intrigue, a cover of Grand Theft Auto V whispering tales of synthetic sin, the undying visage of zombie Jesus feasting upon a Subway offering, a hint of gore enveloping our beloved Shrek, unearthed from the annals of time.
As narrated by the minstrel of MattVidPro AI, the images birthed by the model yearn to be enshrined in the annals of a specific style.
DALL-E 2, a paragon of its era, graced the realms in April of 2022, veiled in secrecy and guarded by the whispers of an unseen waitlist, a conundrum of popularity, ethics, and safety. Yet, the gates swung open in September of that selfsame year, welcoming one and all to partake in the mysteries of the AI image realm.
