Inspiration
I had a song idea that had been brewing in my mind for a long time—a gothic metal piece about unreachable love, set in an underwater world. I wanted to express this through both music and visuals.
But traditionally, creating a music video requires a full band, professional filming equipment, actors, locations, editing teams, and a budget of hundreds of thousands of dollars.
Then I discovered AI tools. Suno for music, Midjourney for characters, KLING for video generation. I had never used any of these tools before, but I had 3 days during a holiday weekend.
I thought: "If I have the vision, the tools, and the time, why not try?"
This project represents not just a finished music video, but proof that AI has democratized creative expression. Solo creators can now realize visions that were once impossible without major studio backing.
What it does
Aquabreath tells the story of a woman trapped underwater, reaching desperately toward the surface where her loved one exists beyond her reach. The music video combines:
NARRATIVE STORYTELLING A cinematic underwater journey following the protagonist in her white gothic dress, expressing longing through visual metaphor.
FULL BAND PERFORMANCE Five AI-generated band members (vocals, guitar, bass, drums, piano) performing the gothic metal composition, integrated seamlessly with the narrative.
BILINGUAL ACCESSIBILITY Japanese vocals preserve the original artistic vision, while English subtitles enable international understanding.
The video demonstrates AI's capability to produce complex, character-consistent music videos with both narrative depth and musical performance—a combination rarely seen in AI-generated content.
How we built it
PHASE 1: MUSIC COMPOSITION (10-15 hours)
- Wrote original Japanese lyrics by hand
- Composed melody using MIDI keyboard and Logic Pro
- Synchronized lyrics to melody (syllable-by-syllable matching for AI accuracy)
- Designed tag structure to guide AI arrangement
- Generated music with Suno using custom melody upload
- Final mixing with stem separation
PHASE 2: VIDEO GENERATION (30-50 hours)
- Created character bibles for 6 characters (5 band members + protagonist)
- Generated each band member individually to maintain consistency
- Difficulty ranking discovered through trial:
- ⭐ Bass (90% success)
- ⭐⭐ Piano (70% success)
- ⭐⭐⭐ Guitar/Vocals (60% success)
- ⭐⭐⭐⭐⭐ Drums (40% success - complex limb coordination)
- Used KLING Image-to-Video for all performance scenes
- Lip sync generation for vocal scenes
PHASE 3: VIDEO EDITING (15-30 hours)
- Multi-track editing (8 video tracks for character layering)
- Speed adjustment to match 175 BPM (most time-consuming)
- Integrated Japanese lyric overlays + English subtitles
- Final compositing and color grading
KEY LEARNING: Set expectations at 80% perfection. AI has limitations, but the goal is artistic completion over technical perfection.
Challenges we ran into
1. DRUM GENERATION NIGHTMARE Drums were by far the most difficult. The complex coordination of four limbs plus facial expressions resulted in a 40% "facial collapse" rate where the character's face would distort during intense movements. Solution: Keep drum shots short and strategic.
2. CHARACTER CONSISTENCY Maintaining the same character appearance across multiple shots was crucial but challenging. Created detailed character bibles with reference images and strict prompt templates for each character.
3. SPEED SYNCHRONIZATION AI-generated performances don't naturally match the song's BPM. Required extensive speed adjustment in post-production to make performances look natural while staying on beat.
4. WATERMARK ISSUES Learned the hard way about tool watermarks. Some editing tools add watermarks that are difficult to remove—had to work around this in the final cut.
5. LIP SYNC PRECISION Getting the vocalist's mouth movements to match the Japanese vocals required multiple attempts and careful timing in editing.
Despite these challenges, completing the project in 3 days proved that AI tools have reached a threshold where solo creators can produce professional-quality music videos.
Accomplishments that we're proud of
- Created a complete music video—music composition to final video—in just 3 days
- Successfully generated and integrated 5 distinct band members with consistent characters
- Achieved seamless fusion of narrative storytelling and band performance
- Learned 3 new AI tools (Suno, Midjourney, KLING) from scratch
- Maintained artistic vision despite technical limitations
- Proved that AI has democratized music video production for solo creators
- Created something I had only imagined but never thought I could realize alone
What we learned
TECHNICAL LESSONS:
- AI tools have different strengths: Midjourney for characters, KLING for motion, Suno for music
- Character consistency requires upfront investment in reference materials
- Generate complex scenes one element at a time, never simultaneously
- Set realistic expectations (80% perfection is achievable, 100% is time-consuming)
- Speed adjustment in post-production is crucial for musical synchronization
CREATIVE LESSONS:
- AI is a creative partner, not just a tool—it suggests possibilities you hadn't imagined
- Limitations can inspire creativity (short drum shots became a stylistic choice)
- The "democratization of creativity" is real—what once required studios is now possible solo
PHILOSOPHICAL LESSONS:
- Perfect is the enemy of done—I completed this because I aimed for "good enough"
- AI hasn't replaced human creativity; it's amplified it
- The most important part isn't the technology—it's having something to express
What's next
- Expand to full-size version (3-4 minutes) using the established workflow
- Apply lessons learned to create more efficient production pipeline
- Explore other music genres and visual styles
- Share the production process through articles and tutorials
- Submit to additional AI film festivals (Reply AI Film Festival in Venice, AI Film Festival Japan)
- Continue pushing the boundaries of what solo creators can achieve with AI tools
Built With
- ai-music-generation
- ai-video-generation
- filmora
- kling
- logic-pro
- midjourney
- music-production
- suno
- video