Artificial intelligence (AI) has brought an onslaught of content, including deepfakes, parodies, visual storyboards, and even real commercials. Companies like Byrna and Baerskin are some that have leaned heavily into AI commercials. Various shows on the History Channel have been using AI content to depict historical events in lieu of produced re-enactments. Then there’s the controversy surrounding Tilly Norwood, a synthetic AI character produced and licensed by Particle6 Productions in the UK. Is it only a matter of time before producers use AI to replace traditional production methods for full feature films?
There are the obvious counter-arguments, such as the AI ‘look’. While that might be an issue for some, it’s not really new. Ever since the dawn of filmmaking, non-traditional production methods have been used. Within the last quarter century, films like Sky Captain and the World of Tomorrow and Final Fantasy: The Spirits Within are probably the closest equivalent to the current AI trend. Sky Captain used real actors, but Final Fantasy was totally synthetic, thanks to motion capture, a lot of 3D animation, and tons of SGI computing horsepower. They may have been groundbreaking for their day, but both were financial failures.
Fast forward to modern artificial intelligence production techniques
Brute force computation on-site has been replaced by logging into various LLMs with as little as a laptop. Meanwhile, some data center is crunching away to generate your prompts. The skills of talented computer animators have been supplanted by your ability is to visualize a shot and describe it in minute detail as a written prompt. However, one key difference in the production of a synthetic film up until now – whether it’s traditional animation, pseudo-realistic, or a hybrid – is that it typically requires a large team. AI-generated content is more conducive to having a single producer working on it. The vision has to be tightly controlled and the prompts kept consistent and specific. It’s harder to achieve good results when the creation of AI content is driven by a decentralized team. So the ‘director’ is often the primary creator of the content.
Let’s set aside the issue of the visual aesthetics of AI. Maybe it will get better and become more acceptable, maybe it won’t. It has clearly advanced in a short period of time. Hopefully, the frequent and unprompted hallucinations will stop happening. It’s hard to say. Personally I don’t think so, because of three factors: 1) AI is inherently ‘stupid’; 2) AI does not understand the real world in terms of physics, perspective, movements, etc; 3) There is no financial incentive for AI developers to make visual content better than it is now. That’s because other applications of AI are far more lucrative. For now, AI image generation is good enough for the majority of content creators and consumers.
Will there be a market for an AI-generated feature film? If you don’t include something done specifically for animation or anything that might only have novelty appeal, then I still think the answer is ‘yes’. However, that’s only true if the story is compelling or possibly if it’s a limited, straight-to-streaming project. Many viewers are fine with AI in its current form. Plus, if we are thinking in terms of Netflix, then there’s a huge potential international audience. While you might look at creating an AI feature film as a ‘cheaper’ way to tell your story and maintain total creative control, it’s not the easy route some might envision. So, is it a feasible option for a producer in the world of indie filmmaking?
Comparing an AI-based feature film project
Over a decade ago I wrote a Film Budgeting Basics post that took a look at the cost of indie filmmaking. If you assume that the film is non-union or produced under a special low-budget carve-out, then it could cost anywhere between $50K and $1 million for a 90-minute feature film.
Sadly, in spite of inflation, the rates outlined in that article are still more or less valid today. There are certainly fluke productions that get made at the bottom end of that range and become wildly successful. Nevertheless, most producers can expect to be closer to the $1 million end. The total time frame is typically nine months to a year. Between production and post, possibly as many as 100 folks have been involved. Filmmaking is team art. It’s rarely the total and complete vision of a single writer or director or producer working entirely on their own.
Let’s compare that to a hypothetical, synthetic 90-minute feature film produced exclusively with current AI tools. A typical dialogue-driven film will have around 1,200 shots, each averaging several seconds. However, the shooting ratio (filmed versus used) is likely between 4:1 and 10:1. Footage for all scenes is usually captured over a 20-30 day stretch on a smaller film, utilizing a handful of locations and sets. Often the editing starts on the first day of the shoot and carries through until everything is delivered. Within that time frame, ten weeks are usually budgeted for the editor and director to ‘lock the cut’ and finish the film creatively.
In the case of AI, these same 1,200 shots have to be precisely described (prompted), previewed, and the best version downloaded. Hallucinations and improper rendering of the prompts are inevitable, which means that there is still a good-to-bad shot ratio that’s similar to real production. Given that ratio, a person working for a solid day will probably average four or five good clips/shots each day. This means it will take a solid year to generate the entire film if you work five days a week. Add more time for potential shot changes and re-renders, which will likely push you into the first part of a second year. That effort is more than a real production, but less than the four years that it took to get Final Fantasy out the door. Trying to divide this into a team approach is likely going to cause additional problems, primarily of consistency.
Budgeting AI production costs
The generation of AI content is not free. There are various deals and subscription plans based on AI credits. Every shot you create – even those that are heavily flawed – will cost you credits. You won’t know what the number of credits are for any given shot until it is created. Depending on the vendor’s plan, the allotment of credits may time out. In other words, if you are halfway through the month and have chewed through all of your allotted credits, you’ll have to wait for the month to roll over before you get a new allocation of credits. Conversely, if you don’t use up all of the credits in that month, you can’t save them to be used in the next month. Right now it’s the wild west when it comes to these various AI plans.
Credits are based on a number of variables, including complexity and length. For instance, if you upload a short, simple office shot with talent and prompt the AI engine to keep it largely the same, but change the background, then the number of credits charged will be relatively low. On the other hand, a completely generated one minute film opener will cost you a lot of credits. In fact, some services limited the length and/or resolution of a shot. Need something longer and you’ll be asked to upgrade the plan.
What about the cast?
We’ve all seen plenty of parody videos that use the likenesses of well-known actors, politicians, or celebrities without their consent. The creators get away with this because it is a parody and protected (at least in the US). But that doesn’t apply to a real film. You don’t want to rely on some arbitrary likeness created by the AI software, because you cannot vouch for the legal use of that persona. Therefore, you do need to hire an actual cast.
You can approach the production in three ways. One way is to simply photograph stills of your cast members in various positions and angles. These would be used as reference frames for the AI clips. Likewise, you can record and upload voice samples. This gives you ‘safe’ training content for the LLMs; however, you will need to work out an NIL contract (name, image, likeness) with each cast member. For example, maybe you have a cast of ten actors and you pay each $1,000 for the rights to use their image and voice in perpetuity as part of the film. That’s $10K for the cast.
The second approach would take this up a notch. You could record the entire script’s dialogue in advance with the real actors and then use that as the real audio. Some of the software options allow for matching the AI renders with lip-synced audio. Essentially this is similar to the production workflow used for animated films. Naturally, this increases the cost for the cast, their time, and studio session time.
The third approach would be to shoot all scenes with the real cast and cheap cameras in general environments. Those clips would be uploaded to the AI system with prompts to replace everything – only using those recordings as a guide. For example, an office background becomes a moody newspaper bullpen. An actress is replaced by her AI self wearing different clothing. The audio might be the real audio or possibly augmented with ADR sessions in post. Essentially this is analogous to the way the film A Scanner Darkly was created in 2006. Remember, that the AI is still recreating the entire shot, so the appearance of your actual cast members will be slightly altered in those rendered images.
Ultimately, this is hard to budget without knowing the exact method to be used. However, since some general production days are required, this becomes the more expensive of these three techniques. The upside is that these recordings can be ‘pre-edited’ to hone the story. Then those edited shots become the basis for [hopefully] a more accurate AI result, with ‘real’ performances and emotion.
Even with AI content, post production is required
You’ve generated all of the shots from detailed, descriptive prompts. If everything comes together as planned, there might only be several weeks of editing for refinement. However, if it doesn’t come together – and many films are truly built in the cutting room – you can assume the full ten weeks, along with more AI content creation to fix gaps.
You often don’t know what works and what doesn’t until you see a cut assembled. It’s at that point you realize which scenes and shots can be removed or rearranged. However, my guess is that you have less wiggle room to adjust the cut than with a standard film. That’s because the bad AI renders might not be useable alternate takes in the same way as an unused take in a regular film.
Color correction might also take longer, because the AI ‘look’ is so baked in. Compared with live action shot on film or in RAW or log using a digital camera, the colorist has much less latitude to creatively impact the aesthetics with tools like DaVinci Resolve or Baselight. If you are trying to shift the grade into a different style, the colorist has to work against the stylized color of those rendered clips.
Audio post rounds it all out. Depending on the digital performances of the synthetic AI ‘actors’, you might have a lot more time in ADR, as I discussed above. After all, there’s no real way to truly ‘direct’ the AI to get the range of emotion human actors would be able to give. Some of that can be mitigated in vocal performance recordings with a real cast at the start (as with animation) or through ADR after the fact. The bottom line is that audio post will likely require more ADR or voice recording time, the same amount or more for sound design, and the same number of mixing hours. The music score can also be created through AI, but it will likely be rather vanilla. You won’t get the same ebb and flow to the score in the way that a music composer would produce.
Closing thoughts
I’ve put together a back-of-the-napkin budget estimate for what a 100% AI-generated feature film might cost (click to enlarge or download the image). Naturally, this is more of an educated guess than anything else, simply because it hasn’t been done yet. Plus the economics of AI are evolving and there are many variables, as I’ve already outlined. Hopefully it’s a realistic starting point without cutting corners.
It should be noted that the biggest line item in my hypothetical AI budget is the cost of a director. However, if you yourself are the person generating the AI content and this is your project, then you are the director. Clearly in that case, it’s not a hard cost. However, it does represent the impact of opportunity-loss, because you would otherwise be doing billable work for other clients as a director. That’s definitely something to consider before jumping in.
I do think at some point we’ll see dramatic feature films created – or at least attempted – using only artificial intelligence as the production method. I’m not optimistic that the process will be worth the effort, nor that it will be financially successful. Naturally, I’m talking about a straightforward, dialogue-driven indie film. If the vision is fantastical or something that would get extraordinarily expensive in sets, VFX, and costumes, then AI can be a legitimate alternative in the future. The potential might seem appealing to some and frightening to others. Regardless, no one should view it as a ‘cheaper alternative’ to working on a real project with a team of creative, like-minded filmmakers.
©2026 Oliver Peters
digitalfilms 