CPE - AI-Generated Video Series Production
Tanner Notch
Count Play Explore logo
CPE — Three-Video AI-Generated Brand Series
For Count Play Explore (CPE), a children's education brand, the goal was to produce a three-video Instagram launch series using a fully AI-generative pipeline. Each video featured a distinct character concept: a giant kid moving through a city, a talking dog with a chalkboard, and a giant baby in a suburban scene, each designed to make math and science feel exciting to families.
Working under a defined paid test scope with a single retainer-track engagement on the line, I built and executed a multi-tool generative workflow that navigated platform limitations, dialogue/lip sync challenges, and multiple client revision rounds, all without traditional filmed talent.
The Final Videos Section:
"Kaiju Girl" IG video
Talking Puppy" IG video
"Giant Baby" IG video
Each video was designed for Instagram vertical format, with native dialogue, brand audio, and motion graphics finishing.
1. Concept Development & Multi-Tool Workflow Planning
The CPE project required navigating three distinct creative concepts (giant kid, talking dog, giant baby) each with unique production challenges: a giant scale character moving through real-world environments, a non-human character delivering brand dialogue with lip sync, and multi-character scene composition with action beats. From the outset, this wasn't a single-tool workflow problem... it required orchestrating multiple AI platforms based on each tool's strengths and limitations.
Each platform brought different strengths:
Flora and Nano Banana for reference image generation, feeding video prompts
Veo 3 for cinematic generations with rich environmental rendering
Kling 3.0 for native audio dialogue generation (critical for the Talking Dog video)
Runway Gen-3 for cleaner motion and inpainting/cleanup
ElevenLabs for voiceover and ambient SFX (crowd chants, background atmosphere)
Flora generation example
Image generated from Flora
Screenshot of the "Giant Baby" Flora project
Final image generated from Flora
From Kling 3.0
2. Visual Generation & Asset Production
With references established, generation began across multiple platforms in parallel. Each video had its own platform strategy based on the specific creative challenge:
Kaiju Girl (giant kid through city): Generated exclusively in Kling 3.0 for dialogue-driven close-ups, and because both Veo and Runway prevent the use of any minors in their generations, which required some clever workarounds. The giant scale required careful prompt architecture to maintain proportional integrity, the model had to read as a giant version of a normal kid, not a small adult or a Disney mountain peak interpretation.
Talking Dog (puppy with chalkboard): This was the most technically challenging video. AI lip sync tools typically don't support animal characters, Kling's dedicated lip sync feature explicitly rejected the dog. The breakthrough came in using Kling 3.0's native audio generation (separate from their lip sync feature), which produced clean dialogue with mouth movements that read as authentic puppy speech. This validated a workflow upgrade I've since applied across other dialogue-heavy projects.
Giant Baby (suburban scene): Required multiple character generations (mom, dad, baby) with consistent likeness across shots, plus action beats with dialogue between characters. This pushed the limits of character consistency, requiring re-generations and selective compositing to maintain visual continuity.
video generated in Kling 3.0
Note: A core insight from this production: AI video tools are not interchangeable. Each has specific strengths around motion, audio, character consistency, and content moderation. Production efficiency comes from knowing when to use which platform — and when to combine outputs from multiple sources.
sample generation from Veo
sample generation from Kling 3.0
Section 3: Audio Production & Lip Sync Solutions
Audio was a major technical focus across all three videos. Each video required clean dialogue (sometimes for non-human characters), branded audio moments, and ambient atmosphere.
Kling 3.0 native audio breakthrough: During the Kaiju Girl production, I tested Kling 3.0's native audio generation against my existing ElevenLabs + post-process lip sync workflow. The result was significantly better than expected — Kling's native audio produced clean dialogue with synchronized mouth movement, without the music/melody injection that Veo's audio sometimes adds. This became the new default workflow for dialogue-heavy AI video work.
the Kling 3.0 output that outperformed the original lip syncing version.
ElevenLabs SFX for crowd chant: The Kaiju Girl video required a crowd chanting "Count Play Explore," something ElevenLabs' voice library doesn't directly support. Generated via ElevenLabs Sound Effects (separate from their voice library), the result was a usable crowd chant that read clearly in context.
ElevenLabs "crowd" audio
ElevenLabs "crowd" audio
Voice processing for "giant kid" effect: Per client direction, the Kaiju Girl character needed to sound booming and large to match her giant size. Built a five-effect audio stack in Premiere (pitch shift, parametric EQ, multiband compressor, chorus, large hall reverb with pre-delay) to create the giant voice effect without leaving the editing environment.
Section 4: Client Iteration & Revision Rounds
The CPE project involved multiple revision rounds with client feedback on specific creative beats, pacing, lip sync quality, voice character, color grading, and scene-specific notes. The first delivery received feedback across all three videos, requiring targeted regenerations rather than full rebuilds.
video generated in Runway
Key revision insights:
Color grading callouts required platform-specific adjustment. Different generations from different platforms had different color profiles requiring unification in Premiere/AE.
Voice character refinement required regenerating ElevenLabs outputs with adjusted voice settings, slower delivery, more natural pacing, less compressed tone.
Lip sync precision required either re-generating with Kling's native audio (Kaiju Girl) or careful manual lip sync work with traditional tools (Talking Dog before Kling native audio breakthrough).
Pacing tightening in After Effects: adjusting cut timing, reducing dead frames, and time-remapping selectively to improve perceived energy.
The revision round was completed within the original scope (no scope creep) by stacking tool-level efficiencies: Kling native audio replaced post-process lip sync, ElevenLabs SFX replaced manual audio layering, and AE time-remapping replaced re-generation for pacing issues.
CPE logo
Like this project
Posted May 13, 2026
Produced a three-video AI-generated Instagram series for Count Play Explore.
Likes
0
Views
3
Timeline
Apr 1, 2026 - May 1, 2026
