What is an AI filmmaking stack?

An AI filmmaking stack is the set of tools that carries a project from idea to delivery — image and video generation, node-based control, hosting and inference, consistency training, editorial, and finishing. It is not one app but a chain, each tool chosen for the job it does best.

Which AI video generation tools do you use in 2026?

I use several depending on the shot — Kling, Veo, and Sora 2 for video generation, and Nano Banana Pro for image work. No single model wins every shot type, so I benchmark them by character, environment, and motion and pick the right one per job rather than defaulting to one.

Do you need to code to build an AI filmmaking stack?

You don't need to code to use most of the stack, but engineering leverage helps. I use Claude Code, the Anthropic API, and MCP tooling to automate repetitive steps and connect tools. It turns a manual workflow into a repeatable pipeline, which matters most at production scale.

How do you choose which AI tool to use for a shot?

Fit for the shot, not novelty. I start from what the shot needs — realism, motion, a specific style, or a locked character — then pick the tool that clears that bar most cheaply. Chasing the newest model is how budgets and schedules quietly fall apart.

← Writing/June 16, 2026·3 min read

The AI Filmmaking Stack in 2026: Tools I Actually Use

A working AI filmmaking stack for 2026, category by category — the real tools I reach for and, more importantly, what each one is actually for.

By Wayan Palmieri

This is the AI filmmaking stack I actually use in 2026 — not a shopping list of everything that exists, but the tools I reach for on real projects and what each is for. The stack is a chain, not a single app: generation, control, inference, consistency, editorial, and finishing, each chosen because it does one job well. The discipline is picking the right tool for the shot, not chasing whatever launched this week.

I led AI production end-to-end at Kartel AI and direct my own generative films and music videos. Everything below earns its place because I ship with it.

The stack, category by category

Category	Tools I use	What it's for
Image & video generation	Kling, Veo, Sora 2, Nano Banana Pro	Creating the actual frames — motion, environments, characters, stills
Node-based control	ComfyUI	Repeatable, parameterized generation you can tune and rerun
Hosting & inference	FAL.ai	Running models at speed and scale without local bottlenecks
Consistency	LoRA training	Locking a character, product, or style across many shots
Editorial	DaVinci Resolve	Cutting generated clips into an actual film — pacing, continuity
Finishing	Adobe After Effects	Compositing, cleanup, effects, and final polish
Agentic & engineering	Claude Code, Anthropic API, MCP	Automating steps and wiring the pipeline together

Image and video generation: Kling, Veo, Sora 2, Nano Banana Pro

This is where frames are born. I keep several models on hand because none of them wins every shot. Kling, Veo, and Sora 2 each have different strengths in motion, realism, and control; Nano Banana Pro is where I do a lot of image work. I benchmark them by shot type — character, environment, motion — and pick per job. Defaulting to one model is how you fight the tool instead of using it.

Node control and inference: ComfyUI and FAL.ai

Generation is only useful if it is repeatable. ComfyUI gives me node-based control — parameterized graphs I can tune, save, and rerun so a shot can be regenerated on demand instead of rerolled from scratch. FAL.ai handles hosting and inference, running models fast enough that iteration does not stall on my local machine. Together they turn a promising demo into something a production schedule can rely on.

Consistency: LoRA training

The hardest problem in generative work is keeping a character, product, or look identical across dozens of shots. LoRA training is my main answer — training the model on a locked reference set so a face, a brand asset, or a style holds from shot to shot. Without it you get seven good clips that clearly do not belong in the same film. This is the difference between a reel and a piece.

Editorial and finishing: DaVinci Resolve and After Effects

Generated clips still have to cut. I bring them into DaVinci Resolve and apply ordinary editorial judgment — pacing, continuity, coverage, grade. Then After Effects for compositing, cleanup, and the effects and polish that make footage read as intentional. This is where a decade in the edit bay matters more than any model, and it is the same craft I describe in how I build an AI-native pipeline.

Agentic and engineering: Claude Code, Anthropic API, MCP

The newest layer is engineering leverage. I use Claude Code, the Anthropic API, and MCP tooling to automate the repetitive parts — batch operations, file wrangling, connecting one tool's output to the next. You don't strictly need this to make a single shot, but at production scale it is what turns a manual workflow into a pipeline a team can run.

The stack is not the point

Tools change fast; the thinking doesn't. Every project I scope starts from the shot — what it needs — and works back to the cheapest tool that clears the bar. That shot-by-shot logic is the same one behind AI-hybrid versus AI-native production. The stack above is just my current answer, and it will look different next year.

If you are assembling a stack for a team or a specific production, I help studios and brands make exactly these choices. For examples of what comes out the other end, see selected AI Artist work.