Ad Buyer Ditches Image Models, Uses Seedance 2.5 for Realistic AI UGC Characters
churchkey · x · 2026-09-28
An ad operator shares his workflow for realistic AI characters: image models look too perfect, while video model Seedance 2.5 produces phone-footage realism.
His process: find a real-looking clip on TikTok or Pinterest, capture the exact frame, have ChatGPT convert it into a JSON prompt covering face, hair, skin, clothes, lighting and camera angle, then paste it into Seedance 2.5 at 4s / 9:16 / 1080p and pause on the most natural moment. He says his own prompt-writing style makes a big difference, but even a simple version works — ideal for批量 producing convincing UGC-style ad creative.
More from Multimodal
- Photoshop beta launches Relight for real-time image relighting — rufusd · 2026-09-28
- User has Opus 5.5 make a space exploration film — every frame and note written in code — prasenx · 2026-09-28
- Dev laments Opus 5.5 can one-shot music videos after building his own pipeline — nptacek · 2026-09-28
- Kling 4.0 Confirmed for October; Flash Model Live Now for Ultra Yearly Subscribers — koltregaskes · 2026-09-28
- Early hands-on with Kling 4.0 Flash: impressive prompt understanding — umesh_ai · 2026-09-28
- KP rehashes WebMCP basics with an explainer video generated by Opus 5.5 — thisiskp_ · 2026-09-28