Cinematic AI Music Video Direction and Production
I directed and produced Porte Italiano as a cinematic AI music video for a client down in Culiacán. Instead of treating the project as disconnected AI shots, I approached it as a narrative short film. The story moves through nightlife, romance, violence, and escape. Throughout the film, recurring characters and controlled environments help maintain visual continuity.
The production used Seedance and MiniMax H3 for different purposes. I reserved Seedance for the most important and visually demanding shots. Meanwhile, MiniMax H3 handled a substantial amount of the production. This approach helped me control AI generation costs without lowering the overall production value.
However, switching between models was only part of the solution. MiniMax H3 can soften detail and visual fidelity when the art direction becomes highly specific. To address that limitation, I developed a specialized generative-video workflow for the project. As a result, I could preserve the painterly, cinematic look established for Porte Italiano. The workflow reduced production costs while keeping the finished film visually consistent.
Blender Previsualization for the Fight Sequence
One of the biggest technical and creative challenges was the extended fight sequence. From the beginning, I wanted it to feel like a continuous one-take action sequence. The corridor fight in Oldboy influenced some of its visual language. Rather than solving everything through AI prompting, I planned the sequence with Blender previsualization. This allowed me to define blocking, camera movement, character placement, and timing before generation.
The Blender previs gave me a controllable 3D environment for designing the sequence before final generation. For example, I could test camera paths, character blocking, staging, composition, pacing, and spatial relationships. That process let me solve filmmaking decisions before spending credits on final AI footage. Afterward, those decisions helped guide the generative-video models.
Using Previs to Reduce AI Generation Costs
Previsualization also became an important way to control production costs. Instead of repeatedly generating expensive shots, I established choreography and cinematography beforehand. Consequently, fewer generations were wasted on incorrect camera movement or action. The process also gave me greater control over complex sequences.
At the same time, the workflow combined 3D previs, AI filmmaking, generative video, local AI models, visual development, and traditional filmmaking techniques. AI remained one component of the production process. More importantly, it supported directing, shot design, and problem solving rather than replacing them.
Local AI Models and ComfyUI Production
Part of the production also ran through local AI workflows using ComfyUI, including MiniMax H3. Running models locally gave me more control over both the pipeline and production costs. In addition, it reduced my dependence on hosted AI platforms and per-generation credit systems.
Local processing offered more than cost savings. With ComfyUI, I had direct control over individual stages of the AI production pipeline. For instance, I could test different approaches and integrate specialized techniques. I could also determine which shots truly required premium cloud models. Therefore, each task could use the most efficient production tool available.
Cinematic References and Visual Development
Several sequences developed from specific cinematic references. For example, the knife sequence began with the aggressive energy of the opening scene of Natural Born Killers. From there, I reworked the concept around the characters and visual world of Porte Italiano.
Likewise, Oldboy influenced the fight sequence in a different way. I was interested in sustained action, lateral movement, and visual choreography. However, those ideas served as starting points rather than templates. They helped shape the camera movement, shot construction, editing rhythm, action, and tone of the sequence.
A Cost-Conscious AI Filmmaking Workflow
Throughout production, I evaluated where higher-cost AI generation actually added value. Seedance was reserved for shots where its capabilities were especially important. In contrast, MiniMax H3 and local ComfyUI workflows handled other parts of the production. Specialized techniques helped maintain the required visual quality across both approaches.
As a result, the workflow balanced AI production cost, visual consistency, cinematic quality, and creative control. Local processing and cloud generation were selected on a shot-by-shot basis. Each choice depended on the best combination of quality, control, and cost.
AI Filmmaking as a Complete Production Workflow
Ultimately, the finished music video combines AI filmmaking, Seedance, MiniMax H3, ComfyUI, local AI models, Blender previsualization, visual development, music video direction, and editing. The goal was never simply to generate attractive imagery. Instead, the goal was to direct the film.
Every stage required intentional decisions about shots, characters, continuity, and visual style. When technical limitations appeared, I developed ways to work around them. At the same time, I managed generation costs throughout production. Most importantly, I selected the right production technique for each individual shot.