Poppy
C4D 2020
An AI Film Production Case Study
The Dandelion Hedgehog
This project documents a complete AI film production pipeline — from script to final edit — using Kling, Veo, Seedance, Runway, Gemini, Nano Banana, Firefly, Claude, Luma and Topaz alongside Photoshop and After Effects.
The Story
Script as the foundation
Before touching any AI tool, the story came first. A full script mapped out shots, pacing, and story beats —
giving every production decision a clear creative anchor.
Finding the Visual Language
Style frame as anchor point
The visual direction began with intentional creative inputs — mood references, thematic keywords, atmospheric direction — refined through generations until a single frame captured exactly the world I was after. That frame became the visual bible for everything that followed.
Meet the Character
Designing the Dandelion Hedgehog
The first major production constraint emerged early. Gen AI tools consistently reverted the Dandelion Hedgehog — a creature whose quills are soft wisps of dandelion seed — back into a standard hedgehog, undermining the core visual concept.
Through a combination of AI tools and manual refinement in Photoshop, I rebuilt the character's appearance from the ground up — developing a no-quill reference version that finally gave me consistent control over future generations.
(Building the no-quill hedgehog — a progression)
These references became the foundation for full 360° model sheets for both versions of the character —
the visual bible for all future generations.
From Sketch to Shot
Multi-AI image development workflow
Each composition started as a rough hand sketch — a fast, lo-fi way to communicate framing and spatial relationships that text prompts alone can't reliably convey. When a first AI attempt missed the mark, a second AI refined the prompt. Reference images grounded the style. No single tool handled everything — each was chosen for what it does best.
(From sketch to final shot — a development progression)
Controlling the Narrative
Keyframe extraction and clip
To take back directorial control, I developed a method for anchoring the beginning and end of each generated clip — creating continuity between shots rather than accepting whatever the AI decided to do next.
This created a bridge between clips, allowing me to stitch together a coherent sequence rather than accepting whatever the AI decided to do next.
Image to Video: Finding the Shot
Generative exploration as a discovery tool
Not every shot starts from a tight brief. For certain scenes I used a more open generative pass —
exploring unexpected compositions and motion possibilities before refining selected frames to match the established visual language.
Exploration and precision work as two distinct phases, not one.
Lighting & Color
Cinematic refinement
Getting a strong composition is only half the work. Several images required lighting direction corrections that standard prompting couldn't reliably fix — redirected across multiple AI-assisted passes until the light matched the scene's logic. Color correction was then done manually to ensure tonal consistency across the full sequence.
(Finding the right light — an adjustment progression)
Enhancing for Film
Resolution and delivery pipeline
Extensive iteration across multiple tools inevitably degrades image quality. The final step brought all footage through AI upscaling and detail enhancement — restoring short film delivery standards. Quality control at the output stage is as important as any creative decision made earlier.
The Edit
Bringing it together
Editing AI-generated footage is a different challenge than traditional editorial work. Each clip arrives with its own lighting variations, motion rhythm, and character inconsistencies — the edit has to work harder to create the illusion of a continuous world. Every cut is a deliberate decision. Pacing, duration, and sequence order were all calibrated to keep the tension alive from the first frame to the last.
