Twarz to jedna z najbardziej intrygujących części ludzkiego ciała. Z jednej strony karykaturzyści potrafią w 5 minut tak wyeksponować charakterystyczne cechy, że od razu widać podobieństwo. Z drugiej jednak bardzo często milimetr w tę czy inną stronę robi dużą różnicę. I dlatego z dużym dystansem podchodziłem do koncepcji, która pojawiła się w mojej głowie, aby "zlecić" Chatowi GPT opisanie twarzy Herberta. No bo niby jak? Postanowiłem jednak spróbować. Wpisałem: "Create a memory portrait of Zbigniew Herbert so that his face can be drawn from it." Jak zwykle otrzymałem całe wypracowanie, więc poprosiłem o wypisanie w punktach. Oto, co otrzymałem:
Well-groomed beard? No ciekawe kiedy... Ale dobra, wpisałem to w Midjourney i otrzymałem to:
Herbert jak malowany przez ślepego. Jedno więc jest pewne - ChatGPT nie jest w stanie stworzyć opisu twarzy, na podstawie którego Midjourney mogłoby wygenerować portret pamięciowy przypominający choć trochę opisywaną postać. Postanowiłem więc spróbować czegoś innego.
Poprosiłem ChatGPT o opisanie idealnej twarzy kobiecej: "Create a description of the ideal woman's face that I can paste as a prompt to the artificial intelligence that generates the graphics." Wygenerowany opis lekko edytowałem i wkleiłem do Midjourney:
Photography portrait with dark background. The face should exhibit a high degree of symmetry, with balanced proportions between the left and right sides. Aim for well-balanced facial features, including a harmonious relationship between the forehead, eyes, nose, cheeks, lips, and chin. The skin should be flawless, smooth, and evenly toned, radiating a healthy and youthful appearance. Create a facial shape that is proportionate and aesthetically pleasing, taking into account the forehead, cheekbones, jawline, and chin. Craft eyes that are vibrant, clear, and captivating, with an appropriate size and shape that complements the overall face. Design eyebrows that are beautifully shaped, neatly groomed, and in harmony with the natural features of the face. Create lips that are well-proportioned, with a defined cupid's bow and an ideal volume, conveying attractiveness and femininity. Ensure the cheekbones are elevated, defined, and add a sculpted dimension to the face. --ar 15:10
Efekt utwierdził mnie tylko w przekonaniu, że im dłuższy opis, tym AI ma większy problem ze stworzeniem czegoś, co by mu odpowiadało. Wydałem więc prostą komendę: Shorter please! Na postawie tego, co napisał mi ChatGPT stworzyłem kolejny prompt:
Photography, dark background, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. --ar 15:10
I tak otrzymałem od Midjourney następujące propozycje:
No całkiem nieźle. Skrócenie opisu ewidentnie dało lepszy efekt. Postanowiłem go więc zostawić. W kolejnych wariantach modyfikowałem tylko szczegóły dotyczące tła i formy.
Digital art, cobalt blue background, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. --ar 15:10
Oil painting, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. --ar 15:10
Oil painting by Courbet, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. --ar 15:10
Analog photography portrait, 30s style, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Analog photography portrait, 40s style, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Analog photography portrait, 50s style, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Analog photography portrait, 60s style, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Analog photography portrait, 70s style, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Analog photography portrait, 80s style, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Analog photography portrait, 90s style, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Photography portrait, cyberpunk style, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Photography of woman in lace blouse in a meadow, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Analog photography of woman in lace blouse in a meadow, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Photography of a longhaired woman in a random place, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Goldenhour photography of a longhaired woman in a random place, a symmetrical woman's face with well-defined cheekbones, almond-shaped eyes with long, thick eyelashes, eyebrows that are neatly arched, a straight and proportionate nose, soft and full lips with a defined cupid's bow, clear and glowing skin, a slender and graceful neck, and a radiant smile that exudes confidence and warmth. Lens 35mm --ar 15:10
Złota godzina jak zwykle efekciarska, więc rezultaty wyszły całkiem całkiem. Postanowiłem jednak sprawdzić, co wyjdzie jeśli zrezygnuję z opisu przygotowanego przez ChatGPT, a zamiast tego użyję po prostu słowa "beautiful".
Goldenhour photography of a beautiful, longhaired woman in a random place, Lens 35mm --ar 15:10
Nie jestem specjalistą od twarzy, ale wydaje mi się, że jakaś różnica jest. Przy okazji naszła mnie myśl, by wydusić z ChataGPT więcej precyzji. Napisałem więc: Be very precise, give proportions. W odpowiedzi otrzymałem dość długi opis, który z wyżej wymienionych powodów musiałem nieco skrócić.
Photography of a longhaired woman in a random place, face with a high level of symmetry, the width of the face approximately equal to the distance between the eyes, distance between the hairline and eyebrows about 1/3 of the total face height, eyes positioned at the center of the face horizontally, with the width of one eye between them, the distance between the bottom of the nose and the chin approximately equal to the distance between the hairline and the bottom of the nose, forehead: The width similar to the width of the jawline, cheekbones: prominent but in proportion to the rest of the face, jawline: defined but not overly sharp or wide, chin: The length about one-third of the total face height, eyes: design bright, expressive eyes with an almond shape, placing them approximately one eye-width apart, eyebrows: well-defined eyebrows that align with the inner corner of the eye, peak slightly past the outer edge of the iris, and taper towards the ends, craft full and shapely lips with a defined cupid's bow, where the upper lip is approximately 1.6 times smaller than the lower lip, high and defined cheekbones that add dimension to the face without appearing exaggerated. Lens 35mm --ar 15:10
Wyszło całkiem nieźle. No może poza jakimiś dziwnymi podziałkami. Uznałem, że "złota godzina powinna rozwiązać ten problem". Zmieniłem więc prompt na:
Goldenhour photography of a longhaired woman in a random place, face with a high level of symmetry, the width of the face approximately equal to the distance between the eyes, distance between the hairline and eyebrows about 1/3 of the total face height, eyes positioned at the center of the face horizontally, with the width of one eye between them, the distance between the bottom of the nose and the chin approximately equal to the distance between the hairline and the bottom of the nose, forehead: The width similar to the width of the jawline, cheekbones: prominent but in proportion to the rest of the face, jawline: defined but not overly sharp or wide, chin: The length about one-third of the total face height, eyes: design bright, expressive eyes with an almond shape, placing them approximately one eye-width apart, eyebrows: well-defined eyebrows that align with the inner corner of the eye, peak slightly past the outer edge of the iris, and taper towards the ends, craft full and shapely lips with a defined cupid's bow, where the upper lip is approximately 1.6 times smaller than the lower lip, high and defined cheekbones that add dimension to the face without appearing exaggerated. Lens 35mm --ar 15:10
Efekt wyszedł zgodny z oczekiwaniami. Żadnych linii, za to miękkie światło zachodzącego słońca. Swoją drogą ciekaw jestem kiedy wizerunki generowane przez AI będą na masową skalę wykorzystywane w reklamie. Pole do kreacji jest ogromne, bo to, co tutaj pokazałem to tylko kropla w morzu możliwości. Dla mnie najciekawsza jest łatwość przeskakiwania między epokami i porównywania ze sobą różnych stylów odpowiadających zmieniającym się dekadom. Lubię takie zestawienia. Poza tym Midjourney już teraz daje możliwość komercyjnego wykorzystania grafik. Jedyna bariera to jakość. Maksymalna ilość pikseli to 1024x1024. Zdecydowanie nie jest to rozdzielność na billboard.
