AI Image Creation Methodology
I've been using AI to create images for our website. When I first started using AI the images were simple and cartoonish, but they have gradually become more and more realistic and artistic. Recently I asked AI: Using the image of this young man (B/W photo of actor Connor Storrie only showing his head, arms and torso), create a beautiful image of the Greek god Dionysius, dancing in ecstasy with a thyrsus (a staff of giant fennel, ivy and grape vines leaves and grapes) in an ancient Greek setting with a temple and olive and fig trees in the background.
AI initially created this image with three arms (probably because I had previously used AI to create images of Shiva, who has more than two arms).
I asked AI to remove the third arm holding the thyrsus but to change nothing else.
I was pleased by the colors and textures in the image, even though I hadn't specified those, and so I then asked AI: Explain how you created this image. From where did you source the elements, such as the color scheme, the fabric color of Dionysius's drape, the textures, the vase, bowl and cloth, the temple structure, the background of rocky hills, the form and position of the legs, the stone floor, the form of the olive and fig trees.
AI's answer:
I'm glad you asked. The image isn't assembled from a library of stock assets or copied from particular artworks. Instead, it is synthesized as a new composition based on patterns learned during training from a very large collection of licensed, publicly available, and human-created images. That means each element is generated rather than retrieved. Here's how the major components arose.
Overall artistic style
The image is an original synthesis influenced by many traditions simultaneously:
19th-century academic history painting
Romantic landscape painting
Classical sculpture and archaeological reconstructions
Modern cinematic historical illustration
Contemporary fantasy realism
Mediterranean landscape painting
Rather than choosing one artist, the model blends characteristics common across those traditions.
Color scheme
The warm palette was inferred from several factors in your prompt:
Dionysus is traditionally associated with grapes, wine, sunlight, fertility, and celebration.
Ancient Greece is often imagined in warm Mediterranean light.
You asked for a "beautiful" image rather than a documentary reconstruction.
The model therefore leaned toward
warm golds
olive greens
deep purples
ochres
cream-colored stone
These are colors frequently associated with Mediterranean environments and classical historical painting.
Purple drapery
The purple cloth was not copied from a specific source.
Purple is probably the strongest traditional color associated with Dionysus because it evokes
wine
grapes
luxury
royal garments
theatrical costume
Historically, Tyrian purple was among the most prestigious dyes in the ancient Mediterranean, so the model naturally associated Dionysus with a rich purple mantle.
Fabric textures
The textures are synthesized.
The folds resemble heavy linen or wool because
Greek himatia often appear this way in sculpture
historical films frequently depict similar fabrics
painters naturally exaggerate folds to create movement
The model generates those folds mathematically rather than selecting them from photographs.
Vase, bowl, and cloth
These were inferred from common iconography.
The model recognized that an ancient Greek sacred setting often contains
kraters
bowls
amphorae
woven cloth
wine vessels
It invented objects consistent with those expectations.
Temple
The temple is likewise synthesized.
Its features resemble combinations of
Doric architecture
Ionic columns
Greek sanctuaries
archaeological reconstructions
No single temple was copied.
The proportions are influenced by the visual language of places like
Delphi
Olympia
Bassae
the Acropolis
without intentionally reproducing any one of them.
Rocky hills
These are a generalized Mediterranean landscape.
The model has learned recurring characteristics of
Greece
western Anatolia
southern Italy
the Aegean islands
which commonly include
limestone cliffs
dry vegetation
olive groves
blue atmospheric haze
It synthesized a landscape consistent with those environments.
Olive trees
The olive trees show characteristics the model has learned are typical:
twisted trunks
silver-green leaves
rounded crowns
irregular branching
Those are generated rather than traced from photographs.
Fig trees
Likewise, the broad leaves and branching pattern come from the model's learned representation of fig trees.
Stone floor
The paving is a generic reconstruction of weathered limestone or marble paving frequently seen in archaeological sites.
It was chosen because it visually reinforces
antiquity
sacred architecture
processional spaces
Position of the legs
This is particularly interesting.
The pose emerged from several influences:
your uploaded photograph
classical contrapposto
dance
ballet
Hellenistic sculpture
Because you requested Dionysus "dancing in ecstasy," the model exaggerated the crossing of the legs into something more graceful and theatrical than historically literal.
One can see echoes of
Praxitelean contrapposto
Hellenistic dance figures
modern fashion photography
all blended together.
Face
The face is strongly influenced by the uploaded model, but idealized.
The model tends to
increase facial symmetry
smooth skin
slightly enlarge the eyes
refine proportions
soften asymmetries
This creates a face that resembles the reference while appearing more archetypal or divine.
Lighting
The lighting is cinematic rather than archaeological.
The model chose
low golden sunlight
strong rim lighting
warm highlights
cool shadows
because that style produces emotional impact and dramatic depth.
Composition
The composition follows classical painting conventions.
Notice the triangular arrangement:
head near the apex
crossing legs
temple balancing the right side
olive tree balancing the left
foreground objects anchoring the bottom
This creates stability despite the dancing pose.
Why it "feels Greek"
Perhaps the most remarkable aspect is that the model has learned statistical associations among thousands of visual cues that humans recognize as "Greek":
olive trees
figs
pale stone
Mediterranean light
mountainous horizons
Doric architecture
purple drapery
grape vines
laurel and ivy
idealized youthful anatomy
None of these were individually specified in your prompt. They emerge because the model has learned that these elements frequently co-occur in depictions of the ancient Greek world.
I think this is one of the strengths of working iteratively, as we've been doing. Your prompts establish a historical and iconographic direction, while the model fills in many culturally consistent details. Then, through successive refinements—such as replacing the straight sword with a Macedonian kopis, removing the saddle and stirrups for Alexander, or correcting Bhairava's cobra and iconography—we move from a broadly evocative image toward one that is both artistically compelling and increasingly faithful to the historical or religious tradition you're exploring.
See Yoga East's website and our other blog posts for more interesting images created with AI.
--Laura Spaulding
No comments:
Post a Comment