AI tools for storytelling
A twelve-week studio on diffusion video, agents, and the open AI pipeline — taught inside PolyU School of Design as SD5976.
Twelve weeks. Twelve briefs.
A lecture hands you the material; the workshop is where you push on it. You ship every week. Final showing is open to the school.
Orientation, the safety contract, your first ComfyUI graph, and the brief. Every Saturday after this ends with something you made.
Diffusion fundamentals. Text-to-image, then image-to-video. First LTX pass. First 3D pass. Make thirty bad things on purpose.
Reference imaging, ControlNet, prompt scaffolding. Move from 'AI image' to 'your image.' Audio joins the pipeline.
MCP servers as the lingua franca. Wire a video pipeline that a model can drive end-to-end.
LLMs in the loop — tokenization, context windows, prompt structure. Wire a multi-step agent that produces a usable artifact.
Multi-agent harnesses for storyboarding, asset generation, and continuity. Critique the agent's work; redirect it.
Audio generation depth — diegetic sound, voice cloning ethics, score scaffolding. Agents take a stab at edits.
Storytelling structures that survive the AI middle. LTX-driven sequences, reference-locked characters, consistency hacks.
3D assets feeding the 2D pipeline. Camera moves with depth. Safety & IP review of every model you've leaned on.
Multi-step pipelines you can run unattended. MCP orchestration. The point where the studio scales beyond what one person can hand-author.
Process writing, README discipline, agent-assisted code review. Public artifacts mean public method.
Final crit. Public showing. Archive your X. Hand the pipeline forward.
- A one-minute character video — yours alone.
- A group video project, shipped with the team.
- A public GitHub repo documenting the pipeline.
- A written reflection — the seams, the dead drafts, the why.
We're open about which tools we use. Tools change cohort-to-cohort; the studio practice doesn't.