Bird Song
13,000 impressions & 300 reactions on LinkedIn
Bird Song is short animation designed to stress test modern generative platforms against the realities of daily AI production for professionals in the creative sector.
The primary challenge was to see beyond industry hype and determine how GPT Image 2 and Seedance 2.0 perform for working creatives requiring character consistency, emotional depth, and reliable production pipelines.
Expression and audio.
Seedance 2.0 proved incredibly capable at handling exaggerated close-ups and emotive expressions. Because it is a multimodal video generator designed to parse complex inputs into motion, it is great for nuanced character work. To match this visual quality, I used Google Gemini 3.1 Flash TTS Preview for expressive, controllable audio.
To maintain lip sync and voice consistency, I drove the Seedance audio using a reference black screen video paired with the Gemini TTS snippet. This forced the AI to anchor its audio generation, avoiding synthetic robotic voices.
Consistency and cost.
The biggest visual challenge was generating multi-character scenes. Too much detail from GPT Image 2 character sheets often confused Seedance 2.0, forcing me to abandon them and work backwards from close-ups. This also revealed that while Nano Banana 2 maintains a better original artistic feel, GPT Image 2 is more consistent for adapting to complex camera angles.
Seedance 2.0 demands high computational power, so I restricted my output to 720p to manage credit costs before upscaling the final footage in Topaz Astra.
Gallery
My Bird Song LinkedIn posts have over 13,000 impressions and around 300 reactions.