Share
Explore
ACO - AI Update 1
March 2025
Current Landscape
The Generative Video AI market is exceptionally competitive right now - significant advances are announced weekly, if not daily and it’s difficult to keep up.
Most press articles focus on publicly accessible tools (Sora, Runway, VEO2) however in their current form these tools are less likely to have substantial, direct impact on ACO members and the demand for Operators.
For now at least, the best performing tools are likely in-house, propriety technology developed with custom pipelines and trained on the back catalogue of a studio’s movies and TV shows, unfortunately due to the closed nature of that technology it’s difficult to get a glimpse of how advanced those tools may be until the studio release footage or a project where it has been used. Given current AI video limitations, adoption and use is currently limited to pre-production phases such as storyboards and previsualisation, or specific production use cases such as de-aging techniques used on the film Here (discussed further down). Again, this is rapidly evolving.
That said, the developments of publicly accessible tools remain good indicators of the direction and speed of progress AI video is heading. These tools will empower more independent creators in the short term and provide a credible challenge in some areas such as music videos, short films and commercials where many of AI video’s current limitations are less of a drawback (such as consistent storytelling). This is already being seen.
For the most part these AI video tools are currently limited to generation of clips short in length, from mere seconds to up to 2 mins depending on the platform. Each platform has their own strengths and weaknesses but broadly the current “leaders” in AI video are below. Included are major Chinese platforms which are leading the field in many areas.
List of current publicly available AI Video Platforms (Alphabetical Order)
UPDATE: In a sign of how quickly things are changing, as I prepare to finalise this update Google Veo2 is gradually opening up for public use on paid plans and the video being generated are the most cinematic and realistic by some margin. In the list of examples at the bottom of this update are a number of videos created with Veo2.
Limitations
Close attention must be paid to the limitations of today’s AI video and the on-going work taking place to lessen those obstacles. These developments most likely telegraph the impact AI video may have have on the ACO and it’s members and when.
Many of the limitations are the same regardless of the model so a broad summary is below.
The implication of this is, unless agreements have been signed such as that between Runway/Lionsgate mentioned below, AI video models cannot ‘officially’ be trained on existing copyrighted material . Any video produced by a model which has been trained without proper permission makes the generated video highly susceptible to legal challenges and may discourage the use of such models for commercial work.
In theory this means that Hollywood studios with substantial catalogues are at a significant advantage, however the prevailing thought within the industry is that most AI video models have simply been trained on copyrighted material without permission. Whether this can be proven and how that may impact their usage is a big unknown.
Industry Impact
The latest stories and AI developments having immediate or short-term impact on the film industry.
Staircase AI Studios.
New AI studio announced this month with the backing of Hollywood talent including Divergent producer Pouya Shahbazian, Black List listed screenwriter Michael Schatz and oscar nominated Pixar designer Teddy Newton among others. Their aim is ‘near-major studio quality releases’
They also announced their first feature film, simultaneously releasing the first 5 minutes which are linked to below. They aim to produce 30 low budget projects in the new 3-4 years.
Announcement:
Website:
First 5 Mins:
Lionsgate + Runway Partnership.
The first major studio to officially announce a collaboration with an AI platform. Specifics of the collaboration and precisely how Lionsgate will be utilising the AI platform are vague, initially focusing on pre and post production, however a significant development from the agreement will be the ability for Runway to legally train AI models on Lionsgate’s back-catalogue, including analysing their cinematography and performances from professional actors. This overcomes one of the biggest hurdles to AI platforms which is access to high quality video imagery from which to train their datasets.
Adobe Commercial Safe Content.
Copyright and IP infringement has potential to slow down use of AI and tangle platforms or creators in legal red-tape. Adobe’s Firefly has responded to this by introducing the concept of ‘commercially safe’ AI video where the AI has been trained exclusively on content Adobe has permission to use. This allows creators to generate, use and monetise video created through the platform eliminating one of the legal concerns faced by other platforms.
Metaphysics & De-Aging Tech
An example of the immediate impact AI can have on production. This Advanced AI de-aging technology from Metaphysic ) was used in the film Here. Not only did it greatly reduce the budget thanks to saving cost on VFX but it also changed how the film was shot on set with actors and crew able to see the AI de-aged output in real-time. A fascinating, detailed write-up is available here...
AI Audio Developments
While not video-related, AI audio is rapidly being adopted by Hollywood studios and proves a case study of how AI development can impact traditional filmmaking and the ways it is being embraced by Hollywood, perhaps indicating a trend that will extend to video generation in the coming years.
Tools like Repeecher () and Flawless () are examples of AI platforms making an impact, including the controversy surrounding The Brutalist. A
Capabilities now offered by AI Audio includes;
Amazon has just this week announced AI dubbing of some of it’s titles also:
)
Artistic Rights Treasury
A positive outcome from these audio advances has been progress achieved in protecting the artistic vision of a performance through Artistic Rights Treasury (A.R.T). This is a framework to ensure that all manipulations are done with a performances consent and is backed by the Screen Actors Guild of America. Little information is available about the specifics of the framework but it could be a model to protect AI manipulated acting performances in the future too, a major concern in the recent union talks.
Technology Advances
The pace of advance in the field makes it almost impossible to keep track, but below are some recent milestones I think are worth highlighting. In addition to the links below, for anyone wishing to keep a finger on the pulse of AI video technology I would also recommend the AI News and AI Blog found on the website Curious Refuge
Google Veo2
I’ve mentioned it above, but the short films and music videos generated with Veo2 demonstrate an ability to mimic photoreal, cinematic imagery in a way that is a dramatic improvement over what has been showcased to date. With this platform now being made available to paid users I anticipate we will be seeing a lot more such videos which will give us an indication of how capable Veo2 is and what progress has been made eliminating some of the limitations highlighted above.
Examples of videos made with VEO2 are included in the videos section further down. Between now and the next update I will explore this more an give a more detailed update.
AI Camera Operating
Cinematographic Camera Diffusion Model.
Link:
The first of three research papers discovered that are specifically attempting to help AI understand cinematic grammar and camera operating. While the full journal is behind a paywall they have published the below telling remarks from feedback apparently received from working crew including Director’s and Cinematographers.
E.T. the Exceptional Trajectories: Text-to-Camera-Trajectory Generation with Character Awareness.
Link:
The target of this second paper is more precise mimicking of camera operating. The abstracts of the paper available without paying are highly technical but it’s clear that the work is explicitly trying to enable text written prompts that replicate camera movement and character awareness of camera movement. Of particularly note is the following section illustrating how such a prompt may be entered.
Camera Ctrl Project.
The third area of research is less ambitious in trying to specifically copy Hollywood film operating, but it is attempting to give AI video users more specific control over the camera and in turn make AI video feel a little more authentic to the types of storytelling we’re familiar with.
OmniHuman-1.
A new tool by TikTok owner ByteDance which enables users to take a still photo of a person and turn it into a brief, full motion and audio video of that person moving and talking. This tool is not intended specifically for creative storytelling but does demonstrate how far AI has come in replicating photorealistic human behaviour given the right input.
Website:
Examples:
Video Showcase
The videos below showcase the cutting edge of AI generated video.
Which Side Are You on? by Ruairi Robinson
Type: Short Film
Description: A sci-fi short film created using Google Veo2. Possibly the most cinematic use of AI video seen to date. An interesting case study as the creator is a filmmaker themselves with over a decade creating cinematic imagery through a combination of VFX and practical cinematography.
‘The Pisanos’ - A Porsche Spec Commercial by László Gaál.
Type: Short Film
Description: A spec commercial created by one of the leading independent AI video specialists László Gaál.
One of the most realistic, real-world examples of AI generated video which attempts to mimic real-world cinematography without many of the sci-fi/fantasy embellishments of many AI videos.I highly recommend watching the additional link that shows all of the full-length clips that were generated in Google’s VEO2 from which the edit was pieced.
Advert:
BTS Breakdown:
‘The Blue Stone - Steve Moore’ by Ruiari Robinson.
Type: Music Video
Description: Photorealistic, cinematic fantasy music video from director and VFX artists Ruairi Robinson the same filmmaker behind ‘Whose Side Are You On?
Dawn Chorus by Ruiari Robinson
Type: Short FIlm
Description: Another fantasy/horror short film from Ruairi Robinson testing the photorealistic capabilities of Google’s Veo 2
‘Secret Santa’ - Coca-Cola
Type: Commercial
Description: Fully AI generated Christmas commercial released by Coca Col last year that generated a lot of pushback and controversy. The spot did however achieve the highest test scores yet for a Coca Cola Christmas advert, a fact that is telling both for how likely companies may be to use this technology going forward, and for audience acceptance of such imagery.
Changan Showcase by László Gaál
Type: Short FIlm
Description: An example of Chinese car company Changan using AI to replace a traditional car advert.
Human Character Testing by Ruiari Robinson
Type: Short Film
Description: László Gaál is doing more interesting work on finding a solution to one of the limits of AI video - it’s struggle in reproducing exactly the same character between clips. The first video is his initial, general attempt of generating vignettes of human characters. The second video explores his use LORA’s (Low-Rank Adaptation) a technique to enable consistent characters across multiple AI video clips.
Human Character Test:
LORA Training:
AI Videos Index
Name
Type
Description
Links
Which Side Are You on?
Short FIlm
A sci-fi short film by Ruairi Robinson created using Google Veo2.
Possibly the most cinematic use of AI video I’ve seen to date.
‘The Pisanos’ - Porsche
Commercial
A spec commercial created by one of the leading independent AI video specialists László Gaál.
In my opinion the most realistic, cinematic example of AI generated video I have seen. I highly recommend watching the additional link that shows all of the full-length clips that were generated in Google’s VEO2 from which the edit was pieced.
The Blue Stone - Steve Moore
Music Video
Photorealistic, cinematic fantasy music video from director and VFX artists Ruairi Robinson
Dawn Chorus
Short Film
Another short film from Ruairi Robinson testing the photorealistic capabilities of Google’s Veo 2
‘Secret Santa’ - Coca-cola
Commercial
Fully AI generated Christmas commercial released by Coca Cola.
Changan
Commerical
An example of Chinese car company Changan using AI to replace a traditional car advert.
Human Character Testing
Short FIlm
László Gaál is doing more interesting work on producing realistic human characters as demonstrated by these two links.
The second video is an evolution with a focus on using LORA’s (Low-Rank Adaptation) to enable consistent characters across multiple AI video clips.
Humans Test:
Consistent Character Test:
There are no rows in this table
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
CtrlP
) instead.