Article originally published in the FICAM magazine “Actuality"
Righting by Fabien Marguillard
Picture from Steve Johnson on Unsplash
Artificial Intelligence (AI) is a ubiquitous subject in the media, conversations, and minds; this phenomenon is explained by the innovative nature of this technology, but even more so by its potentially disruptive impact on certain professions and professional organizations.
This meeting was held on April 4th as a moment for exchange and discussion to assess the role of AI-powered software tools in the audiovisual (AV) processes of member companies of FICAM.
AI can be characterized as a sub-domain of computer engineering, which enhances performance and extends the capabilities of software tools in numerous fields, while exploring new creative application domains with generative AI.
AI-powered software tools (software/AI) have been notably used in the fields of VFX and animation for many years. It is the highly publicized emergence of new generative AI tools like ChatGPT or Midjourney that has sparked many fears.
The report titled “What impact does AI have on the film, audiovisual, and video game industries?” recently published by the CNC distinguishes:
Four important legal points
How secure and confidential is the data provided?
Legal protection of created works
Postproduction
AI is present in software solutions for video coding conversion to HDR (plugin) and upscaling (from HD to 4k). It is integrated into Resolve software for image tracking.
For sound, it is used for reducing audio signal noise.
For voice: some uses of voice morphing and ADR, to generate the voice of a known singer over an actor’s performance; to create a draft of a voice-over for a sound editing guide (actor’s agreement for the final version?); to simulate the expected voice for future ADR, with the synthetic voice being replaced later by the actor’s voice recording for the final version.
Localisation
AI has been used for about fifteen years for linguistic transcriptions and translations intended for program localizations. This automation, complemented by a new phase of work known as post-editing, results in productivity gains, cost reductions – making localization accessible to new programs – and without a negative impact on service providers’ revenue.
Automatic translations are sometimes trained on proprietary datasets. Some clients question the use of synthetic voice generators, and some are waiting for a legal framework to be defined. Some companies conduct technological watch or R&D on generative solutions for synthetic voices.
Currently, in the absence of a legal framework for the origin of training data or for the use of generated content, producer clients examine contracts and question service providers about their internal processes (copyrights, GDPR, TPN…). Voice recordings made in the studio may be intended for generative AI training datasets. The practice of so-called full-buy-out contracts, granting unlimited exploitation rights of a recording for video games and audiovisual content, could be practiced in some countries.
Adaptation - Doubling
Voice actors express their concerns publicly regarding the potential competition from generative AI systems for synthetic voices. An exceptional meeting of the Dubbing Monitoring Committee will be held via video conference on April 22, with the signatories of the DADR agreement, and the participation of about thirty representatives from stakeholders in this activity.
Currently, generative voice AI is not used for fiction because the quality of interpretation is not deemed satisfactory by clients. However, producers are interested in creating synthetic voices through generative AI, which could be a cost-saving measure. These solutions are being tested, and their technical progress improves the quality of the created electronic voices. Experiments are being conducted in language dubbing to enable the distribution of programs in new countries.
Internationally, dubbing unions want to add a specific clause to copyright transfer agreements to prohibit the use of their voice recordings for AI training purposes: there is a risk of dubbing strikes in Spain and Italy if agreements are not reached with the commissioners.
Technical service providers are caught between the protectionist efforts of actors and the desire of producers to experiment with generative voice solutions. They do not want to be held responsible for the use made by their clients of the voice recordings they commission. There is a risk of new international actors entering the market, potentially engaging in unfair competition via the internet by bypassing local obligations.
In France, the visually impaired do not want synthetic voices to be used for audio description of programs. AI-generated SME subtitles are imperfect, especially concerning color code compliance.
Image tricking
AI outperforms 3D for face replacement/aging/rejuvenation treatments. The cost of face-swapping has been reduced by 50 in 10 years. Rejuvenation can be processed either by a generic AI without training or by a personalized AI with training based on historical archival images of the actor.
What rights apply in this case if film excerpts are used? AI is integrated into software to improve automatic masking functions for still images (Photoshop) and animated images (rotoscoping); and to enhance motion tracking and calculation of on-screen motion data (match-moving).
Service providers are questioning their right to use professional VFX software/AI tools that are marketed while being trained with data of undetermined origin. The European AI Act regulation will be transposed in France and will require software/generative AI editors to declare the sources of their training datasets. Some clients believe that visual effects can be produced automatically by software/AI tools, and they would like to minimize the costs of the services provided.
Administrative, financial, legal and operational management
With the example of Moviechainer, a French tool for financial management of program exploitation rights chain. The CNC has authorized the editor BackUp System to use RCA data to train Moviechainer’s AI, which ensures optical recognition and classification of information elements appearing in contractual documents.
Formation
Stéphane Bedin discusses the strategic challenges of professional training, which must adapt to the new paradigms induced by the integration of AI into the information systems of all sectors. He recalls that in the context of a recent report, the government encourages structuring of higher education offerings and an increase in continuing education in AI tools.
In real-time video production activities as well as in post-production, operators are required to indirectly use AI, either in integrated forms within specific software functionalities or through plugins. For some technicians, accommodating AI will involve knowing how to phrase a prompt to articulate their needs precisely.
The "Artificial Intelligence Commission" submitted its report entitled “AI, an ambition for France” to the French President on March 13.
The “Artificial Intelligence Commission” presented its report titled “AI: an ambition for France” to the President of the Republic on March 13th. This report reflects a contradiction between the need to open access to databases that could train French AI solutions and the desire to protect the rights of authors and creators.
Some of its recommendations are considered vague. Didier Huck confirms that during discussions with Anne Bouverot (co-chair of the commission), he identified a risk of overly strong legal protection of French content, which could weaken France’s position in the upcoming international competition in the field of AI: how to ensure that French works are present in learning databases while protecting the rights of their creators? An international conference could be organized on this topic with the participation of the same French specialists after the deadline of the American presidential election in February 2025. Other bodies for reflection and decision-making should emerge at both the national and European levels.
The AI observatory set up by the CNC for the audiovisual and cinematic industries is intended to continue its work. Working groups could be formed at FICAM to monitor the effects of the emergence of AI in AV processes and professions.