Pinned Tweet
FIRST PASS film/video editing building in Codex for DaVinci Resolve.
Not an auto-editor.
I’m trying to make it reason like a real post-production assistant: ingest, selects, string-outs, continuity, multicam, pacing, sound, codecs, color management, Fusion, QC and delivery.
Not far fetched.
my spidey senses are telling me the labs are, as we speak, using unguardrailed versions of their unreleased models to systematically reverse-engineer ALL of the internet’s software, apps, and games, then feeding the reconstructed source code straight into their training pipelines while simultaneously building guardrails to ladder-pull the rest of us
Wind on location will ruin a perfect interview faster than anything. A dead cat on the shotgun and finding the leeward side of a building are worth more than any plugin later, and they cost you two minutes.
Before generating your AI video/film, create a character and location reference sheet. Think of reference sheets as your visual anchors: they help the AI remember who’s on screen and where the scene takes place.
It prevents AI from inventing characters/locations/scenes.
A 2D my daughter vs father kung fu fight in Nairobi City animation.
All tools used in @runwayml: GPT Image 2 for character and location reference sheet generation, Seedance 2.5 for image-to-video generation, and Lyria 3 Pro for sound generation.
Prompt refiner: ChatGPT.
The battle // WITHIN
Tools // Gemini (Image creation) + @Kling_ai (Photo to Video) + Davinci Resolve (Edit, Color Grade, Subtitles and Sound design)
Sound Bed // HARMLET (performed at the National Theatre - UNITED KINGDOM) + Music (FOURTH OF JULY - SUFJAN STEVENS)
Consent conversations go better when you show people the kind of photos you take before you take any. It gives them something concrete to say yes or no to, rather than a vague idea of a stranger with a camera.
Grok Bot is so addictive (in a great way), my bottle neck is my brain churning out ideas quickly and consistently.
Thread on running a small documentary crew in remote places, from what fifteen years across East Africa and the Sahel have taught me about keeping things calm.
Agree roles before you arrive. One person leads conversations with the community, one handles sound, and everyone knows who decides when to stop rolling. Confusion travels fast in a new place.
// ARM WICK //
A lone, relentless fighter tears through a Westlands nightclub in Nairobi to reach the crime boss waiting in his glass-walled VIP lounge.
With the city glowing beyond the windows, every fight brings him closer to a confrontation he may not walk away from.
When First Pass over-selects clips, it's usually because my brief was unsure. Tightening the brief helped far more than raising the clip quota, which is a lesson that applies to human assistants too.
A 2D my daughter vs father kung fu fight in Nairobi City animation.
All tools used in @runwayml: GPT Image 2 for character and location reference sheet generation, Seedance 2.5 for image-to-video generation, and Lyria 3 Pro for sound generation.
Prompt refiner: ChatGPT.
A humanitarian story usually needs fewer photos than the brief asks for. Ten pictures that move from place to person to moment will hold attention longer than forty that all say the same thing in slightly different light.
Motion graphics in documentary work should stay quiet. A clean lower third and a simple map do more for the viewer than animated flourishes, which tend to distract from the people you're trying to show.
A good interview question for impact films is often just asking someone to describe an ordinary morning. The details they choose tell you what changed, and they say it in their own words rather than the programme's language.




