Dreamsplice

Inspiration

I’ve loved Japanese anime since childhood—particularly the cinematic realism and emotional rhythm of Akira, Ghost in the Shell, and Cowboy Bebop.
When I started learning AI animation through Framer’s Cartoon Heroes course in August 2025, I realized I could finally blend those cinematic principles with emerging generative tools.
The idea behind Dreamsplice came from a simple question:

What if dreams could be recorded like film—and edited until reality breaks?

That thought evolved into a story about memory, guilt, and control, expressed through two characters: Riku Arata and Yuna Kaede.

What it does

Dreamsplice : Hollow Code is a short AI-animated film that tells the story of Riku, a young man haunted by his own digital reflection, and Yuna, a hacker who tries to stop him from rewriting reality.
Technically, the project demonstrates how a solo creator can direct an anime-level short by combining:

• AI-assisted storyboarding and cinematography.
  
• Consistent character generation using my 5-Step ANIME ENGINE framework.
  
• Hybrid editing (AI + traditional post) for fluid continuity.
  

How we built it

1. Pre-Production – Wrote a concise 3-act outline, defined emotional color logic
   [(\text{Red} = \text{Danger},\; \text{Cyan} = \text{Dream},\; \text{Orange} = \text{Reality})].
   
2. Character Design – Created Riku and Yuna using consistent DNA prompts:
   hair color, lighting tone, and emotional default.
   
3. Scene Generation – Built key frames in Grok Imagine and Runway, testing continuity every 6 seconds rather than rendering the full storyboard at once.
   
4. Post-Production – Assembled scenes in DaVinci Resolve, added analog film grain, color grading, and ambient rain FX.
   
5. Sound Design – Mixed retro synth with real rain recordings to achieve 1980s acoustic warmth.
   

Challenges we ran into

• Raccord & Continuity: AI struggled with smooth transitions between shots; we solved it by inserting short close-ups to “bridge” motion.
  
• Model Inconsistency: No single model handled every camera move, so we used a rotating model roster.
  
• Iteration Fatigue: Each client-style revision meant retesting lighting and palette logic.
  
• Compute Cost: High-resolution frames required batching and post-upscaling to stay within GPU limits.
  

Accomplishments that we're proud of

• Achieved third place in a Japanese AI-animation contest—my first professional recognition and income from AI animation.
  
• Built a functioning AI anime production pipeline that other creators can reuse.
  
• Proved that a solo filmmaker can reach near-studio quality through structured pre-production and disciplined prompting.
  

What we learned

1. AI isn’t a shortcut—it’s a collaborator. The director still controls rhythm, tone, and symbolism.
   
2. Cinematography is language. Learning raccord and shot logic turned random frames into storytelling.
   
3. Hybrid thinking wins. Knowing both AI and traditional post yields a larger solution space.
   
4. Consistency beats novelty. Locked character DNA and lighting rules are worth more than new effects.
   
5. Continuous R&D is survival. Models evolve weekly; stable pipelines prevent chaos.
   

Mathematically, the creative cycle felt like an iterative function:
[ f_{n+1} = f_n + \Delta(\text{learning}) ] —each scene refinement fed the next until convergence on emotional clarity.

What's next for Dreamsplice

• Extend the world into a five-minute sequel, exploring Yuna’s perspective.
  
• Publish a free AI Anime Director’s Guide based on the 5-Step ANIME ENGINE framework.
  
• Launch a community challenge to help freelancers transition from AI-toy users to production artists.
  
• Continue collaborating with Framer to integrate real-time cinematography tools for AI filmmakers.
  

Ultimately, Dreamsplice proved that structured creativity and AI can coexist—that one person, a clear workflow, and the right direction can build an entire cinematic universe.

Built With

• sora2