Short Reel

In May 2026, I had 48 hours to respond to an open brief: create a short video for Instagram Reels, TikTok, or YouTube Shorts. I chose to build the piece around UGC's narrative and visual conventions: direct-to-camera delivery, approachable language, and a social-native pace. I handled the entire process, from the idea and character design to generation, editing, and motion graphics. The challenge was to coordinate three decisions: speak to a specific audience, capture attention with fast hooks, and keep an AI-generated protagonist recognizable across different scenes.

The project began as a technical test for an AI Content Creator role. The brief left the topic and treatment open, with one condition: deliver a short piece for vertical social media. My goal was to turn that freedom into a proposal shaped by a clear audience, narrative, and audiovisual production rationale. I led the direction and end-to-end production of the piece. I defined the concept, designed the character and visual system, directed the generative iterations, and built the final result through selection, editing, sound design, and motion graphics.

# Shifting from introspection to direct contact. I explored an emotional, introspective story, but discarded it because it required a slower build and made it harder to introduce fast hooks from the outset. I reframed the proposal around direct-to-camera delivery and UGC narrative conventions.

I defined natural features to avoid a generic digital celebrity and achieve a presence compatible with the spontaneous tone of the format.

I used the brand color in clothing and objects as a visual continuity resource, without making it the center of each shot. The character sheet specified a sweatshirt in `#CC484C`, which helped maintain this criterion between scenes.

CONCEPT AND VISUAL SYSTEM I used ChatGPT to explore the narrative and develop prompts. I created the protagonist with Higgsfield Soul 2.0, locked in her features through a character sheet, and generated the still scenes with Nano Banana Pro. VOICE AND MOVEMENT I produced voice variants with ElevenLabs and compared video models using the same prompts. Kling 3.0 was the primary model; Kling 2.6, Google Veo 3.1, and Seedance 2.0 were used for comparison. SOUND AND EDITING I generated background music with Suno, focusing on rhythm, energy, and the audience's age profile. I selected the version that supported the edit without competing with the voice and finished the piece in DaVinci Resolve. DELIVERY I adjusted the final export bitrate to eliminate visual glitches and achieve stable playback.

Character consistency was the main visual challenge. The early attempts retained the overall idea but changed facial features or clothing between generations. To reduce this variation, I created a character sheet with defined views and attributes, then reused it as a reference for subsequent scenes. For the audio, I compared multiple options until I found a voice that felt less commercial and closer to the everyday language of social media.

In a new iteration, I would also evaluate models released after this production, including Seedance 2.0 with 4K output and Gemini Omni Flash. I would not adopt them simply because they are new; I would compare their character consistency, editing capabilities, and fit for the final cut against the original workflow. Because the test required a single piece, the process was designed as a manual production rather than an automated workflow. If the goal were to develop a campaign or a reusable template, I would propose a modular system from the outset: structured prompts, reusable character and prop sheets, shot-by-shot traceability, file-naming conventions, batch generation, and quality controls before editing.