The tools behind the work

These tools are not interchangeable. Each one is better than the rest at something, and most projects pass through several. Here is what we use, and for which job.

What we use, by job

The AI video tools we use, grouped by the job they do
Job	Tools
Video generation	Sora — long, dense shots with complicated motion Google Veo — photoreal frames and believable camera movement Runway — tight control over a shot, and fast iteration Kling — human motion and faces, where other models stumble Luma Dream Machine — flowing camera moves and transitions Pika — short effect shots and quick variants Higgsfield — camera-move presets for ad shots Hailuo — stylised frames and fast bulk generation
Images and references	Midjourney — style references and opening frames Nano Banana — precise image edits that hold a character Flux — product images, and text inside the frame Ideogram — typography that survives generation Adobe Firefly — commercially clean licensing where that matters
Voice and audio	ElevenLabs — voice in most languages, cloning included Suno — original music when a library track will not do Udio — a second opinion on music and style variants
Avatars and lip sync	HeyGen — presenters, and the same script in several languages Synthesia — training video with a corporate avatar Hedra — expressive facial animation from a single image
Edit and finishing	Adobe Premiere Pro — the main edit After Effects — graphics, titles and frame repair DaVinci Resolve — grade, and holding one palette across shots CapCut — fast vertical cuts and captions Descript — editing by transcript, and cleaning voice Topaz Video AI — resolution and frame-rate recovery
Script and planning	Claude — scripts, hook variants and content plans ChatGPT — ideas, adaptations and alternate lines Gemini — reference research and pulling material together

How a tool gets picked for a shot

The shot decides, not habit. A person talking goes to one model, a landscape with a camera move to another, and a product that has to stay exactly itself goes to whichever one holds detail without inventing any. Most finished pieces have shots from three or four different models in them, matched in the grade so the seams do not show.

This list changes every few months, and that is the point — loyalty to one tool is how you end up fighting it. What does not change is that somebody has to sit and watch the seams.

The stages these sit inside: how AI video is made. What came out of them: the work.

Questions about the tools

Does it matter which model a video was made with?

Less than the people using it. Models change every few months and each is better at something — one at motion, another at faces, another at holding a product consistent across shots. Picking per shot is the job; loyalty to one tool is how you end up fighting it.

Why not just use one tool for everything?

Because no single tool is good at every stage. Generation, upscaling, voice, edit and grade are five different problems, and the output of one has to survive the next. The stack matters less than the fact that somebody is watching the seams.

What we use, by job

How a tool gets picked for a shot

Questions about the tools

Wondering whether your idea generates well?