Inspiration
After creating Monster, I wanted to challenge myself with something completely different. Instead of cinematic realism, I felt drawn toward something cute, handcrafted, playful, and slightly eerie.
Around that time, Halloween had just passed. The streets were still covered in leftover decorations, half-carved pumpkins, and fading holiday colors. That transitional mood sparked the question:
What happens to a little demon girl after Halloween, when the world stops celebrating?
I began experimenting with various handcrafted aesthetics — paper cutout, clay animation, textile art — but the patchwork / stitched-fabric look felt the most charming. It carried the softness, loneliness, and surreal dream-logic I wanted for the story.
From there, After Halloween emerged naturally: a small demon girl wandering a patchwork city that shrinks, expands, and bends to her imagination — or perhaps to her loneliness.
What it does
After Halloween is a stop-motion-inspired AI music video blending Kawaii Bass, J-Pop phrasing, and a storybook-horror visual tone.
The narrative follows NYARAÉ, a little demon girl who appears to be lost in a human town after Halloween ends. But the visuals suggest something else:
- A train ticket she throws away
- A portal door she sealed herself
- A city that swells and collapses like a dream
- A world made of fabric, felt, buttons, and clay
- A parade of playful creatures singing and dancing with her throughout the night
The MV intentionally blurs the line between:
- Is she lost?
- Or does she simply not want to go home?
It embraces a sense of cute loneliness, wrapped in tactile, handcrafted textures reminiscent of children’s storybooks and stop-motion films. The project is part of the Dissonant Labs series, exploring emotional storytelling through hybrid AI techniques.
How It Was Built
Music & Vocals
- Composed in Suno v5, using rapid-fire Japanese phrasing, piano ornamentation, and layered Kawaii Bass drops.
- Vocal structure mixes spoken intros, whispered transitions, and energetic melodic choruses.
- The arrangement balances playful mischief with sudden moments of sadness, echoing the emotional arc of the story.
Visual Aesthetic
- Character and environment design created with Midjourney v7 and Google Nano Banana, focusing on patchwork textures, stitched surfaces, and cloth-like lighting.
- Google Nano Banana and Seedream 4.0 were used to generate multi-angle, consistent images of NYARAÉ and key environments.
Higgsfield Suite (Sora2), Wan 2.2 (local runtime), and Midjourney v7 provided motion generation, stylized stop-motion movement, and controlled camera-path tests.
This hybrid workflow allowed me to combine subtle stop-motion qualities with smoother, animation-style motion, including:
- slight frame irregularities when needed
- gentle fabric flicker that enhances the handcrafted look
- a patchwork world presented with a sense of real 3D spatial depth
Animation & Editing
- Edited and composited in Wondershare Filmora 14.
Additional animation techniques included:
- manually adjusting clip speed to match dance rhythm, lyric timing, and approximate lip movement
- manual cut-outs for transitional beats
Production Time
Approximately 30 hours from early experimentation to final render.
Tools Used
- Suno v5 – Music composition & vocal generation
- Midjourney v7 – Character and environment design
- Google Nano Banana – Multi-angle consistency generation
- Higgsfield (Sora2) / Wan 2.2 (local runtime) / Midjourney v7 – Video synthesis & motion generation
- Wondershare Filmora 14 – Editing, pacing, post-processing
Challenges
1. Maintaining Character Consistency in Patchwork Style
Because the character’s facial features are extremely small, even slight variations in I2V generation caused distortion. Achieving consistent expressions and materials required careful image selection and multiple generation attempts.
2. Achieving Cohesive Patchwork-Style Environments
Each scene required well-crafted prompts and accurate reference images to maintain a balance of Halloween aesthetics, cuteness, and subtle darkness. I also needed to consider the character’s distance from the camera, correct lighting direction, and appropriate scale within each handcrafted environment. Scene generation demanded significant attention to detail.
3. Creating Lip-Sync Without Dedicated Tools
Since I chose a stop-motion-style mouth design and could not rely on paid lip-sync services, every lip-sync moment was achieved through editing alone — matching near-correct mouth shapes with the song through manual timing.
Accomplishments that I'm proud of
I’m proud that I was able to complete this entire project. During the early and middle stages, After Halloween actually caused a lot of frustration — the animations generated by models like Kling and Hailuo were far from what I envisioned, and I struggled to find the right ideas for storyboards and scene design.
Fortunately, after spending a lot of time thinking and experimenting, a few “breakthrough moments” allowed the project to finally move forward. Wan 2.2 also turned out to be unexpectedly perfect for this visual style, which made it possible for the final version to take shape the way I imagined.
Finally, I’m proud that After Halloween represents a creative shift in my work: moving from high-energy, human-realistic action (like Monster) into stylized, handcrafted fantasy. It pushed me to explore a new artistic language while still staying true to the emotional tone I want for NYARAÉ’s universe.
What I Learned
This project deepened my understanding of:
- handcrafted-world storytelling
- emotional pacing in music-video format
- balancing cuteness and eeriness within a single visual language
Most importantly, I realized that things which appear simple on the surface are often the most challenging to create.
What's next for After Halloween
I don’t have a specific direction in mind yet, but I’m considering exploring a “giant monster” concept — something in the spirit of Pacific Rim. The idea of using AI to create massive kaiju battling with raw power — without needing the Hollywood budget of tens of millions of dollars — is honestly exciting. If AI can bring that scale and intensity to life, how can you not get fired up?
Built With
- filmora14
- google-banana
- midjourney
- sora2
- wan2.2

Log in or sign up for Devpost to join the conversation.