GPT-tweaked portrait + structured SD prompt turns a selfie into a gritty Ronaldo favela football clip
techhalla · x · 2026-09-12
The author first used GPT to tweak a selfie into a convincing "best Ronaldo ever" portrait, then fed a structured prompt to an SD video model to generate a 30-second vertical clip. The prompt has three blocks:
- STYLE + CAMERA + ATMOSPHERE: raw mobile footage look — handheld shake, grain, compression blocks, slight overexposure, lens dirt, autofocus hunting, motion blur.
- Scene: a gritty concrete football court in a Brazilian favela at dusk, with worn painted lines, chain-link fences, laundry lines, crowd phones half-raised, plus detailed audio cues (ball thuds, Portuguese shouts, screams on the goal).
- IMAGE REFERENCES: the provided photo acts as the single strict lock for the protagonist across the whole take for character consistency.
It's a reusable photo-lock + cinema-verite prompt workflow for AI-generated realistic sports clips.
More from Multimodal
- YuE2 Open-Sources Frontier Music Model with Symbolic Planning and Agentic Editing — multimodal-art-projection · 2026-09-12
- AI Video Rethink: Generate a Strong Draft, Then Keep Shaping It — AIwithGhotai · 2026-09-12
- Nano Banana Pro prompt for vintage ink architectural sketch style — SimplyAnnisa · 2026-09-12
- Seedance 2.5 + CapCut PC: a creator workflow for AI-made product commercials — ZabihullahAtal · 2026-09-12
- MiniMax RefMod ComfyUI workflows: reusable identities without training — Hearmeman98 · 2026-09-12
- One AI video ad workflow: GPT-6 Astra for 3D, Seedance 2.5 for video, CapCut for edits — nikola_mr64990 · 2026-09-12