MT. Iturbide
I named a mountain after myself: Stable Diffusion with ControlNet's Illusion Diffusion shaped terrain from a roadside photo in Mexico into my own face, expanded with Generative Fill and animated with 2.5D parallax.
Final Piece
2.5D parallax camera drift, rotoscoped wildlife, fog layers, and a mountain that resolves into a face at distance. Seven stages, one piece.
Conceptual Framework
Personal & Cultural Origin
A photo from a roadside in Mexico, land tied to family, heritage, and the Iturbide name. A face in that terrain makes the mountain a self-portrait and the title literal: MT. Iturbide.
Generative AI as Craft Tool
Generative AI is one instrument here, not a shortcut. The outscaled periphery and the face-embedded mountain both needed manual refinement, color matching, and animation work. Intent and execution stay human-directed.
Process Breakdown
Source Photograph
One iPhone photo taken roadside in Mexico, looking over a green hillside from a limestone outcrop. The raw 4x5 capture supplies the terrain, lighting, and composition everything else builds on.
Live Photo Capture
Live Photo mode caught a few seconds of ambient motion: wind through grass, shifting leaves, drifting clouds. That became the motion reference for the final scene.
Generative Outfill
Adobe Generative Fill expanded the 4x5 frame into a much wider canvas, keeping the grain, lighting angle, and botanical detail of the source. Passes with different seeds were blended to hide the seams.
Living Elements
Rotoscoped wildlife went into the landscape: birds crossing the frame, insects, ambient animal movement. Each element was color graded, motion tracked, and placed in its parallax depth layer.
Likeness Source
A self-portrait from a cornfield in Mexico gives ControlNet's Illusion Diffusion the facial geometry to encode into the mountain. A face rooted in the same landscape the composite depicts.
Illusion Diffusion
ControlNet's Illusion Diffusion uses the portrait as a conditioning map. The prompt asks for “rocky mountain terrain” and the network bends shadows, ridgelines, and highlights to the face's contours. At the right distance the terrain resolves into a face.
Technical Deep Dive
Illusion Diffusion Architecture
ControlNet takes the likeness as a structural depth map that constrains diffusion: terrain is generated, but light, shadow, and geology must follow the face's alpha mask. Many iterations were compared to balance plausibility against recognition.
Seamless Generative Outscaling
Outscaling meant matching the source's noise profile, film grain, and lens distortion across generated pixels. Passes with different seeds were blended at the boundaries to remove seams.
2.5D Camera Projection
The flat composite was cut into depth layers: foreground rocks, midground vegetation, background mountain, sky dome. Displaced in 3D space in After Effects, they give a drifting virtual camera real parallax.
Atmospheric Compositing
Fractal noise layers roll fog at several altitude bands. Light rays and particle dust follow one lighting direction across every layer, and the wildlife is graded to sit inside the stack.
