Research Context: Constructing Stylised Characters Through Hybrid AI Assisted and Manual Topology Workflows
My character creation pipeline emerged from a hybrid space that sits between traditional digital sculpting and newer AI assisted mesh generation tools. The images show the iterative mesh deformation process where I used AI outputs as reference guides rather than as final assets. This aligns with current discussions on AI as a previsualisation tool rather than a replacement for the artist. Researchers in procedural and AI augmented modelling often emphasise that AI generated references can accelerate ideation but still require extensive human intervention to achieve production-ready topology. My workflow reflected this directly. The AI provided a stylised face for the girl that I then had to manually reshape, remodel and simplify for animation purposes.
Blender’s face generation plugin served as a bridge between the sketch and the sculpting environment. It produced a quad based template mesh, but the responsibility for appealing design and correct deformation topology remained with me. This mirrors findings across digital character research that argue for the primacy of topology literacy. A mesh can only deform as well as its underlying loops allow, and stylised faces exaggerate this challenge because even a small distortion breaks the illusion. My early tests demonstrate this clearly. Shifting the camera angle caused the AI inspired mesh to collapse visually. The face looked unintended and warped when lit from certain angles. This problem is frequently cited in stylised character design: the sculpt must work from all angles, not just the front.
Blindness, Silhouette Logic and Symbolic Costuming
Designing the girl with a blindfold introduced a layer of symbolic logic that is consistent with research on costuming in animation. Scholars note that clothing is often used not simply to dress a character but to signal narrative constraints, psychological conditions or mythic archetypes. The blindfold provided an instantly recognisable silhouette and also reduced the complexity of rigging the eyes, which supported the painterly abstraction that defines the film. The two strands of hair framing her face created a readable visual rhythm and prevented the silhouette from flattening. Choosing to reduce the polygon count after the initial high poly sculpt aligned with best practice in stylised animation pipelines. Rendering constraints, simulation overhead and the need for rapid iteration required a mesh that balanced detail with efficiency.
Facial Identity, Photo Reference and Manual Corrective Sculpting
The father’s design followed a different path. I used my own face as a structural reference, which aligns with historical practices in animation such as rotoscope inspired tracing, performance capture reference and identity mapping. In stylised CG, artists often rely on elements of their own facial geometry to produce anatomy that feels grounded. The father’s long nose, heavy brow and tired expression were deliberate reversals of the friendly father archetype in children’s media. The scars, asymmetric folds and downturned lips follow documented techniques in production design where damage and age are communicated through low frequency deformities rather than high frequency details like pores.
The plugin that allowed me to pin my facial features onto the mesh became less useful over time. It was too restrictive to adapt to the stylisation I required. This resonates with research on procedural rigging and auto adaptive meshes. Automatic systems accelerate early formation, but hand sculpting remains essential for style dependent adjustments. The decision to animate manually rather than rely on pinned reference points reflects this limitation. It also reinforced my understanding that stylised motion requires custom sculptural logic rather than pure automation.
Stylisation Through Deformation and Painterly Surface Integration
The screenshots showing the progressive distortion of both characters situate the work within contemporary debates on painterly rendering in CG. A key problem in painterly animation research is keeping faces legible when detail is intentionally flattened or abstracted. My characters had to work under a brush stroke filter that constantly reinterpreted surfaces. This placed unusual pressure on the underlying forms. The silhouettes needed to communicate personality even when texture information was unstable. Character researchers often highlight stylised proportioning as a solution to this and my work mirrors that approach. The father has a long rigid skull structure. The girl has wide orbital spacing and a soft mandible. Large readable shapes survive painterly abstraction in a way subtle details cannot.