01 / 04    Scroll ↓

Multi-disciplinaryAI creator

I direct commercials, short films and UGC with generative models — concept to final grade. Locked character references first, then plates, then motion, with custom models trained in ComfyUI when a stock one will not hold the character. Same face from the first frame to the last.

Commercials

01 / 04

Crocs — Desert

01

Product spot built on landscape and scale. No dialogue, no captions.

Seedance 2.5, Nano Banana Pro

Evening Arrival

02

Night city, one figure, one arrival. Mood before message.

Seedance 2.5, Nano Banana Pro

Short Films

02 / 04

Doorway

01

Thirty seconds of narrative built entirely from generated shots. One character, continuous light, no cuts that break the space.

Seedance 2.5, Premiere

Deadline — Episode One

02

A capybara sits on a sofa in a frantic open-plan office. An exhausted worker notices her, sits down, and stops. 30 seconds, 12 shots, two characters held consistent across every cut.

Seedance 2.5, Nano Banana Pro, Gemini Omni Flash, CapCut

Kira Infinite

03

A thirty-second game announcement built as one unbroken take. Character sheet locked first, then location plates, then motion — same face and same weapon from the first frame to the end card.

Seedance 2.5, Nano Banana Pro, Higgsfield

Fashion

03 / 04
Sophia persona on a night street with a drink01 / 08

@sophia_travels — AI Influencer

01

One persona, eight settings — night street, overcast, harsh sun, beach, interior, sunset. Skin texture and identity held across every frame, product placement integrated.

Nano Banana Pro, Higgsfield

Mirror

02

Activewear piece shot as a phone mirror take. Fabric and fit read at full resolution.

Kling, Nano Banana Pro

UGC & Presenters

04 / 04

Vitamin C Serum

01

Talking-head review, product in frame from the first second, burned-in captions.

Seedance 2.5, GPT Image 2, ElevenLabs, CapCut

Lymphatic Drainage

02

Hook-led ad. The claim lands in the first two seconds, the demo carries the rest.

Kling, Nano Banana Pro, CapCut

Leave-In Conditioner

03

Longer-form presenter spot for a haircare launch. Warm grade, single take feel.

Kling, Nano Banana Pro, CapCut

Sophia persona on a night street with a drink01 / 08

@soffiatravels — AI Influencer

04

One persona, eight settings — night street, overcast, harsh sun, beach, interior, sunset. Skin texture and identity held across every frame, product placement integrated.

Nano Banana Pro, Higgsfield

Toolkit

Tools & project types
I specialise in

Video models

  • Seedance 2.5
  • Kling
  • Kling Motion Control
  • Google Veo 3.1
  • Sora 2
  • Wan 2.7
  • Minimax Hailuo 2.3
  • Minimax H3
  • Flux.3 Video
  • Grok Imagine 1.5
  • HappyHorse
  • Luma
  • LTX-Video
  • Runway
  • Veo

Image models

  • Nano Banana Pro
  • GPT Image 2
  • Midjourney
  • Flux
  • Gemini
  • Higgsfield Soul 2.0
  • Higgsfield Soul Cinema
  • Seedream 5.0 Pro
  • Grok Imagine 2.0
  • Z-Image
  • Topaz
  • Krea 2
  • Ideogram
  • Qwen Image

Custom models

  • ComfyUI
  • LoRA training
  • Model fine-tuning
  • SDXL
  • Flux
  • Custom workflow building
  • Dataset preparation

Techniques

  • First & last frame control
  • Motion transfer
  • Lipsync
  • Character & face swap
  • Inpaint & relight
  • Reframe and aspect conversion
  • Upscale & restore
  • Cinematic camera control

Audio

  • ElevenLabs
  • AI voiceover
  • Sound design
  • Lip-sync
  • Diegetic sound build

Post

  • CapCut
  • Premiere
  • Colour grading
  • Motion design
  • DaVinci Resolve

Project types

  • Commercials
  • Short films
  • UGC & talking-head ads
  • Vertical drama
  • Fashion & lookbook
  • Product stills
  • Character design
  • Storyboarding
  • Social-first vertical
  • AI influencer build & management
  • Music video
  • Affiliate & performance video
  • E-commerce product ads

Craft

  • Character consistency
  • Hook & retention writing
  • Artefact repair
  • Shot lists
  • Brand guideline work
  • Spatial blocking
  • Reference pipelines
  • Prompt debugging

About

I work shot by shot: locked character references first, then plates, then motion. Most shots land within three or four iterations, and the ones that do not get diagnosed rather than regenerated blindly. When a stock model will not hold a character across a campaign, I build the pipeline in ComfyUI and train the model myself.

English C1. Working remotely across time zones.

Got a project in mind?

Let’s connect