This is how we make films with AI at Talentless AI, written for people making films without a studio behind them. It covers how to decide what to make with AI, how the process works from story to final cut, which tools are worth knowing, and the rules you need to understand before you sell anything.
For most of film history, the size of your story was limited by the size of your budget. If you couldn't afford a period street, a storm or a crowd, you wrote around it.
AI takes away a lot of that limit. A small team can now put things on screen that used to need a studio, including locations that no longer exist, big weather, crowds and entire worlds.
It doesn't make the film good on its own. You still need a story worth telling, someone directing it, and an editor with taste. Those parts take as much time and judgment as they ever did.
When someone brings us a script and a budget, the first real question is what's ready for AI and what isn't. That depends on what you actually want to shoot and how much money and time you have. Most projects fall into one of three paths.
The smallest budget usually means a fully generated feature. Before you commit to that, make one scene and see if you're happy with it.
Don't prompt that scene yourself. Prompting well takes a lot of practice, and whatever you come up with on a first try won't show you what the tools can really do. Pay someone who does this every day to make the test scene, and judge the approach on that.
If you know you want real actors, start mapping out how much of the film you can shoot on gray screen and what you want the world to look like. That's what makes the rest possible, and it's covered in its own chapter below.
Then run a test. Take your casting tapes, or any footage you've already shot, and try to build a scene around it. Making one scene is the first step on this path too.
If you have the money and the time, make a full pre-visualization of the whole movie in AI before you shoot anything. You end up with a rough draft of the entire film to work from, so every creative decision after that comes from something you can actually watch.
This is roughly how our own films came together, including Live in Reverse and Double Wide Desires. Every project is different, but the order holds up.
Both films started with a lot of thinking, writing and outlining before anything was generated. AI can help with that, and the next chapter covers how, but you need to know what the film is about before you start making pictures.
Next you work out who's in the film and where it happens. Each character gets a character sheet showing their face from several angles and what they wear. These sheets are what keep a character looking like the same person from the first scene to the last.
Then you decide what each location looks like and whether it needs to be locked down in 3D. Locations that have to stay exactly the same across many shots can be locked in Blender, a 3D tool VFX artists have used for years. With an AI assistant connected to Blender, the AI can do much of the building. The finished set can be rendered from any angle, and those renders become the references that keep the location consistent. A Blender render can also be the starting point for a shot that gets turned into finished video with a video-to-video tool. This works for hybrid films as well as fully generated ones.
Get the characters and locations approved before you generate any shots. On client work we always get sign-off at this stage. On bigger projects it's also worth approving a still keyframe for each shot before turning it into video. On my own projects I often go straight to video, but the characters and locations are always settled first.
Then the prompting starts. It's a bit of a crapshoot, but it's also an iterative process that gets better as you go. For each section of the film, I generate everything on the shot list, then keep exploring to see what else comes out. I keep going until I'm happy with every shot, the same way a director on a set does as many takes as it takes.
I'm not an editor, so I work with one. I hand over the shots for a section, the editor cuts it and comes back with the gaps, and I generate whatever's missing. That back-and-forth continues until the film works.
On Live in Reverse and Double Wide Desires, all of the sound was AI. The music, the voices and the effects came from ElevenLabs.
I'm a writer, and I don't have rules about where AI is allowed in my writing. That's up to each person.
What I've found is that AI is a harsh critic and a great brainstorming tool. It's very good at rewriting, which means it's good at iteration, at repurposing material, at coming up with new angles and at finding plot holes.
I've tried writing natively in AI, and it's not good at dialogue. Can it write a script from start to finish? Yes. I just haven't found one I'm happy with creatively. I wish it were a better writer, but it's still a tool designed by engineers, so I think writers' jobs are safe.
There's a footnote that matters more for independent filmmakers. You don't have to start with a script anymore. You can write with video, by making scenes and finding the story as you go.
If you plan to sell your script, know the rules. In the US, text generated only from prompts isn't protected by copyright, and WGA contracts say AI can't be credited as a writer and AI output doesn't count as source material.
Pre-visualization used to mean storyboards, or a rough animated version of a few hard scenes. With AI, you can make a rough version of your entire movie instead.
How far you take it depends on your budget and your time, and on where you'd rather spend them. A single scene might be enough to prove the approach. A whole-film pre-vis costs more, but it gives you and everyone you work with a draft of the movie to react to before anyone shows up on set.
The same work helps you raise money. A pitch trailer or a finished proof-of-concept scene shows investors the film instead of asking them to imagine it. We make pitch trailers. Be clear with investors about what's generated and what will actually be shot.
If you're shooting real actors and building the world with AI, gray screen gives you much more flexibility than green screen. It's also where we're hoping to get a lot more experience.
Green screen comes with rules. Your actors can't wear green, the lighting has to be precisely right, and all you're really doing is filling in a background plate behind them.
With gray screen and generative compositing, the AI fills in the entire world. You can change what your actors are wearing after the fact, add props, and turn a table into a castle. The things on set become stand-ins for whatever the scene needs. You still want something real anywhere an actor touches or holds it, because performances suffer in empty space.
You shoot with flat, even lighting, which means fewer setups and more work done in a day. The look gets decided later, because tools like Beeble can relight the whole scene, so you can shoot in the daytime and turn it into night.
Get written consent from every actor whose face, body or voice you'll change, before the shoot. More on hybrid AI production.
These are grouped by what they're for. The list changes often, so we keep specs and version numbers on the models page, where every claim is dated and sourced.
WRITING AND RESEARCH
These are good critics and brainstorming partners, and they're strong at rewriting.
Claude (Anthropic) · ChatGPT (OpenAI) · Gemini (Google) · NotebookLM (Google, for research across your own documents)
IMAGES, CHARACTERS AND LOOK DEVELOPMENT
Use these to build character sheets, locations and the look of the film before you make any video.
Midjourney (strongest for style and texture) · Nano Banana Pro (Google) · GPT Image (OpenAI) · FLUX (Black Forest Labs) · Seedream (ByteDance)
3D SETS
Blender is a free 3D tool that VFX artists have used for years. With an AI assistant connected to it, you can build a set and render it from any angle.
Blender (free and open source)
VIDEO GENERATION
Each one is good at something different. Our models page has a dated, sourced comparison.
Seedance (ByteDance) · Kling (Kuaishou) · Gemini Omni Flash (Google DeepMind) · MiniMax (MiniMax, including open weights) · Luma Ray (Luma, with EXR output) · Runway (Runway) · Midjourney video (Midjourney) · FLUX Video (Black Forest Labs)
CHANGING REAL FOOTAGE (HYBRID)
These start from video you shot and change what's around the actors.
Beeble (generative compositing and relighting) · Runway Aleph (edits footage from instructions) · Luma Modify Video (restyles while keeping the performance)
PLATFORMS THAT BUNDLE MODELS
These put many models under one login, which helps when you're comparing them or building a pipeline.
Flora (node-based canvas) · Higgsfield (camera moves and effects) · OpenArt (image and video models, character consistency) · Dreamina (ByteDance, runs Seedance) · Beeble Canvas (node-based compositing)
VOICE, MUSIC AND SOUND
Everything on Live in Reverse and Double Wide Desires, including music, voices and effects, came from ElevenLabs.
ElevenLabs (voice, sound effects and music) · Suno (music)
FACES AND LIPSYNC
Use these only with the consent of the person whose face or voice it is.
HeyGen (avatars, lipsync and translation)
FINISHING
These are the less exciting tools that make it look like a film.
Topaz (upscaling and cleanup) · Magnific (upscaling) · DaVinci Resolve (edit, grade, and its own AI tools) · Premiere Pro (edit)
Nobody pays to be listed here.
Software is the smallest cost. Subscriptions and generation credits add up, but they're minor compared with the cost of people's time.
On a fully generated film, most of the money goes to the weeks spent directing, generating, choosing and editing. On a hybrid film, you add a shoot, though it's usually shorter and simpler than a traditional one.
How many credits you'll burn depends on how many options you want for each shot. Plan for misses, because not every generation works the first time and every retry costs credits. Spending a little on a test scene first tells you a lot about what the whole film will cost.
People tend to skip this part and regret it later. We checked everything below on 18 October 2026, but these rules change, so check again before relying on them. This isn't legal advice, and you should talk to an entertainment lawyer before you sell anything.
Copyright. The US Copyright Office's January 2025 report says copyright needs human authorship and that prompts alone aren't enough. Human selection, arrangement and changes to AI material can be protected, and it's decided case by case. Keep records of what you wrote, shot, chose and changed.
Writers (WGA). The 2026 WGA contract, ratified this year and running to May 2030, keeps the AI protections from 2023. AI can't write or rewrite literary material, AI output isn't source material, writers can choose to use AI with the company's consent, and a company can't require it. Studios also have to notify the Guild if they license writers' work to train AI.
Actors (SAG-AFTRA). The 2026 TV/Theatrical contract builds on the 2023 rule that a digital replica of a performer needs their consent and payment. If you're working with union actors, read the AI sections before you scan anyone.
Consent outside the unions. Don't use a real person's face, body or voice without written permission, whether or not they're in a union.
The Oscars. The Academy's rules say AI tools neither help nor hurt a film's chances, and that voters weigh how much a human was at the heart of the creative work.
Festivals. Every festival is different, and many haven't written rules yet. Sundance asks nonfiction applicants about AI use and how they'll disclose it. Cannes reportedly bars films where generative AI drives the script, visuals or main performances from its Official Competition. AI-focused festivals, including the Austin AI Film Fest, welcome AI work. Read each festival's current rules, and disclose how you made your film.
Tool terms. Read the commercial terms of every tool you use. Some open models restrict commercial use in some countries.
The set keeps getting emptier. Backgrounds went first, then skies and weather, and now costumes and props can be changed after the shoot. We're heading toward filming actors in plain clothes on a gray stage and deciding everything else later.
That moves a lot of prep into post. Design choices that used to be locked months before the shoot can now be changed after it, and tools VFX artists have used for years, like Blender, are becoming part of how AI films get made.
The tools are also moving from short clips toward whole scenes. In about a year, they went from a few seconds to twenty-second takes, and keeping a long sequence consistent is the problem everyone is working on now.
All of this makes the performance more important. If everything around the actor can be built later, what you're really capturing on the day is the acting.
Yes. You can make a fully generated feature, shoot real actors and build the world around them, or make a traditional film with AI in post. Whichever you choose, make one scene first and see if you're happy with it before committing.
Start with one scene. If you're going full AI, pay someone experienced to make a test scene from your script. If you're shooting real actors, take footage you already have, even casting tapes, and try to build a scene around it.
For a test that decides how you'll make your film, no. Prompting well takes practice, and a first attempt usually won't show you what the tools can actually do. Bring in someone who does this every day.
That's up to you. AI is a strong critic and brainstorming partner, and it's good at rewriting and finding plot holes. It's weak at dialogue. If you plan to sell the script, keep in mind that under current US Copyright Office guidance, text generated from prompts alone isn't protected by copyright.
Green screen limits what actors can wear, needs precise lighting, and only replaces the background. On gray screen with flat lighting, generative compositing can build the whole world, change costumes, add props and relight the scene later. You need fewer setups, so you get more done in a day.
It's a rough version of your film made with AI before production. Instead of storyboards, you can make a draft of the whole movie, so every creative decision builds off something you can actually watch.
You can build a set in Blender, often with an AI assistant doing the building, and render it from different angles. Those renders become reference images that keep a location consistent, or the base for a video-to-video shot.
The parts a human authored can be protected. The US Copyright Office's January 2025 report says human authorship is required, prompts alone aren't enough, and human selection, arrangement or modification of AI material can be protected. It's decided case by case, so talk to an entertainment lawyer before you sell.
The Academy's rules say AI tools neither help nor hurt a film's chances, and that voters consider how much a human was at the heart of the creative work. The Academy can also ask how AI was used.
It depends on the festival, and the rules change every season. Sundance asks nonfiction applicants about generative AI use and disclosure. Cannes reportedly bars films where generative AI drives the script, visuals or main performances from its Official Competition. There are also festivals built for AI films, including the Austin AI Film Fest. Check each festival's current rules before you submit.
Only with their written consent, and with payment if they're union. SAG-AFTRA contracts require consent and compensation for digital replicas, and doing it without consent is a problem whether or not anyone is in a union.
None of them is best at everything, and they change monthly. We keep a dated comparison on our models page.
SOURCES
Send us your script or some footage, and we'll talk about what's ready for AI. We're in Austin.
Let's talk