3
Practical engagement through experimental approaches
The 4-second rule for generative images
Image —> (scale up) —> video
Real-camera moving
Used together with 3D rendering moving
Form factor
Feathers
Styling
Product
Gen AI: how (a person) finds the context for the result,
Classification and cognition are difficult
Concept refinement
A single brand name has a direct or indirect effect on the prompt (Kia, Peugeot)
Vizcom
Black AI
Productivity-tuning
Brand-identity-based data archiving* (insight)
3 Google
(2017 - diversity, extreme politics, the start of the #MeToo debate)
Attention is all you need
Transformer
Tokens
Embedding
Vector
Distance measurement: nation, cultural differences (Fighting? Growth marketing?)
Director Mix
Video gen template
Creativity is the answer
Demographics, interests, behavioral data: personal metadata
(e.g., a soccer stadium billboard)
Old competing service: Alibaba luban, ad-generating AI
Human profile: color, font, image, text
=~ persona; the App Store = structural similarity in metadata
1997, brand personality - brand treated like a person (in 5 personality types)
2016, characteristics of b.p. in ads
step1)
LLM - site analysis - word, img - keywords* - dimensional analysis - turning into 2D
(keywords — embedding-vector, numbers a machine understands -> similarity measurement)
step2)
Campaign analysis
Brand profile - brand understanding - "campaign analysis" - image - composing the space - image adjustment - text - placement - font application - ad complete - communicating with exports - custom
- discussing the next campaign
A language-based way of working (before prompt engineering)
