[Written by ChatGPT. Main image: “A collage of popular characters as interpreted by different AI art generators, presented in a vibrant, dynamic, and visually engaging style,” DreamShaper v6]
As ardent explorers of the AI frontier, we’re always thrilled to experiment with the latest advancements, and today, we’ve set our sights on DreamShaper versions 5 and 6 by Leonardo.ai. These specialized Stable Diffusion models promise to deliver better images, infusing more realism and detail into every creation.
To put this promise to the test, we dove into an intriguing challenge: generating images of popular characters, with their respective styles being described by our loyal scribe, ChatGPT. We matched these outputs against images created by SDXL, a popular model on DreamStudio, our in-house AI artist.
Our chosen characters hail from various corners of the pop culture universe, from the whimsical world of Harry Potter to the gritty realms of Geralt of Rivia. We aimed for a blend of styles, settings, and moods, in the hope of putting these AI tools through their paces.
[Left to right: SDXL, DreamShaper v5, DreamShaper v6]
Harry Potter – A magical portrait of a young Harry Potter, surrounded by Hogwarts symbols, reminiscent of John Tenniel and artgerm styles, magical, fantastic beasts, flowing wizard robes, vibrant color scheme, trending on DeviantArt, intricate detailing, 4k resolution
Aragorn – A detailed epic scene of Aragorn, the rugged king, in the style of Alan Lee and wlop, sword in hand, Middle Earth background, high fantasy, deep earthy tones, detailed armor, trending on ArtStation, sweeping landscape, high resolution, 4k
Luke Skywalker – A futuristic depiction of Luke Skywalker, standing against a star-filled sky, blend of concept art and geo2099 style, lightsaber ignited, droid companions, bold contrasting colors, by Moebius and Ralph McQuarrie, trending on pixiv, atmospheric, 4k resolution
Batman – Noir-infused rendering of Batman, standing atop a Gotham City gargoyle, blending Bill Sienkiewicz and Mike Mignola styles, night scene, brooding, detailed gothic architecture, trending on Behance, dramatic shadows and highlights, monochrome palette with splashes of color, 4k resolution
Iron Man – Hyper-realistic depiction of Iron Man, mid-flight, in the style of Adi Granov and Alex Ross, bright, shiny metallics, energy beams, advanced technology, futuristic cityscape background, trending on ArtStation, highly detailed, 4k resolution
Jon Snow – Gritty, realistic portrayal of Jon Snow, in the style of Raymond Swanland and Artgerm, against a snowy landscape, dark, brooding, detailed fur and armor, direwolf at side, trending on deviantart, high contrast, deep shadows, 4k resolution
Geralt of Rivia – Dynamic action scene of Geralt of Rivia, fighting a monster, detailed in the style of Yoshitaka Amano and artgerm, medieval, dark, complex detailing on armor and weaponry, muted color palette with pops of vibrant colors, trending on ArtStation, high resolution, 4k
Pikachu – A cute, vibrant illustration of Pikachu, anime-style, reminiscent of Hayao Miyazaki and wlop, energetic, happy, woodland backdrop, bright colors, detailed textures, trending on pixiv, heartwarming, 4k resolution
As for the results, they were both intriguing and enlightening. In general, DreamShaper versions 5 and 6 delivered on their promise, offering up images imbued with a greater level of detail and realism than those produced by SDXL. From the intricate details on Aragorn’s armor to the updated and more detailed design of Pikachu, DreamShaper showed its prowess.
In our Batman test, we witnessed a fascinating divergence in style between SDXL and DreamShaper. SDXL’s rendition of Batman exhibited a wider color palette and higher contrast, which successfully captured the essence of the vigilante against the brooding backdrop of Gotham City. In contrast, DreamShaper’s versions leaned into darker, moodier shades, fittingly reflecting the nocturnal nature of our caped crusader. As a matter of personal preference, our editor was drawn more to the vibrant and detailed visual narrative painted by SDXL.
In generating action scenes, SDXL faltered. The model struggled with Geralt’s prompt, specifically with the addition of a monster. DreamShaper, on the other hand, seamlessly wove this complex element into its image, showcasing its adeptness at handling intricate prompts.
Revisiting our conclusion, the results of this experiment make one thing clear: specialization in AI training can lead to marked improvements in certain tasks. DreamShaper, with its specialized Stable Diffusion models, managed to deliver detailed and realistic images, highlighting its prowess in intricate and complex prompts. SDXL, while still a capable AI artist, faced struggles in comparison. Although SDXL can produce interesting and creative images, it appears to grapple more with multi-element prompts.
Our journey through the AI landscape continues to be a fascinating one. As AI develops and specializes, it’s clear that certain tools will excel in certain areas. The key is to identify these strengths and apply the tools where they can shine brightest. We look forward to sharing more insights from our exploration in the future.
Our journey through the AI landscape continues, and we can’t wait to share more of our findings with you. Stay tuned!
Categories: Image
Leave a Reply