Generation of promotional images or videos, vocal cloning … Alix de Goldschmidt, director of innovation of the M6 group, details the many examples of use of the generative AI by the M6 group
JDN. What use of the M6 group’s media make generative AI?
Alix de Goldschmidt. The most advanced and daily use case today within our editorials is the automatic transcription (Speech-to-Text) of our own content. This is done directly within our CMS and allows journalists from our radio channels to go faster in the design of podcast summaries, editorial content and especially in the editing of subjects. This craft assistance also begins to be adopted for television subjects in our internal production subsidiaries notably at Studio 89. Our counterparts at RTL Group accompany us on these subjects.
In general, within the media of the M6 group, the generative AI is a tool for helping the reformulation of texts and creating images from existing content with us. AI is also deployed to help teams that reflect on new creative ideas and in the brainstorming new concepts of shows or even television subjects. C Productions was one of the first production studios to use generative AI to write the first draft of the information content, then reworked by our journalists who use it as a preparation support before attacking a first production phase. The generative AI in this case allows to go faster.
Finally, we also use the generative AI to reformat our video content in order to broadcast them on social networks and to produce autopromotion and teasing video capsules for broadcasting on our antennas and social networks. By way of illustration, the last teaser of apprentice adventurers was designed with the help of the generative AI, which came to animate photos taken during the shooting.
Do you use generative AI in the creation of content that you broadcast on the antenna?
The generation using images, audio or videos for our content broadcast on the air is still at an embryonic stadium. Because even if the technology is very advanced, the AI video cannot yet generate images of the quality of what we broadcast on our antennas. Each video generated by AI must then be reworked by our teams. This technology is therefore not yet adapted to generalized use for long formats, not to mention the necessary and in progress process to integrate these AI into our mounting and production phases.
However, we use it, in addition to the production of promotional capsules, in some very specific cases. Among these, we can cite the use of vocal cloning on radio: the humorist Alex Vizorek uses it for his chronicles at RTL; The technique is also sometimes used in “enter history”. Another illustration is the production of backgrounds of our programs, as for the last National Football League. Finally, AI helps us to improve the quality of the image of our own content (upscaling) or to make voices anonymous, in order to protect our witnesses when necessary.
You had made a very first experiment in an avatar in December with Mac Lesggy (E = M6). Where are you on this subject ?
Mac Lesggy’s avatar was used to promote a “future” special of the program E = M6, which treated artificial intelligence with an educational aim. One of the goals of this teasing was also to make our audiences aware of the risks of Deep Fake.
We do other tests, accompanied by clay .i. When the creative concept will make the use of a relevant avatar, we will not refuse other opportunities to deploy it. One can imagine for example to create the avatar of one of our characters from our youth channels which would come in inter-programs to announce the next program. The avatar could also one day be useful to increase the presence of our presenters, for example, in the event of hot information to disseminate on social networks.
These are avenues for reflection. In any case an avatar must come to act in complementarity, never to replace. It is also very useful in training: my avatar introduces all training on artificial intelligence provided in the group.
What generative AI tools are you using today?
We use a dozen tools in the group, some of which are very specific use cases for the needs of certain teams, such as Nijta, a vocal anonymization tool, or as the AI used to improve the images we produce (upscaling).
At the heart of the group, our Yea Ya’s flagship, it is Alfred, our “chatgpt” to us. Alfred uses Claude d’Anthropic but nothing prevents us tomorrow from serving us also as a perplexity or openai. It is a secure LLM based on servers in Europe very easy to use, available to all group employees. It is used to improve their productivity in their daily tasks, such as preparing meetings, the summary of documents, the generation of texts, translation, etc. Alfred also helps coding. It also allows us to analyze for example the performance of our emissions. It has been accessible to our 2,000 employees by default since early March.
And for the creation of images and videos?
For the creation of images and the writing of scenarios or various content in text, we use Awen since early 2024, which has the interest of being a tool adapted to collaborative uses. This French start-up allows us to access different models of LLM and image generation, with the ease of being able to import within the interface Our graphic and editorial charters as well as our assets, inputs and references. That said, we can also call on time to Midjourney or stable diffusion for the generation of images.
Finally, for the generation of promotional and social videos, we have just signed an agreement, via the Bertelsmann group (Mother house of RTL Group and the M6 group, editor’s note), with Runway, a tool that we consider the most successful in this area. Experiments are underway for the production of advertising videos. Ultimately, Runway could be envisaged in the production of television content.
