Créer une chanson avec l'IA Gemini (Tuto Lyria 3 Pro)

This video introduces Lyria, a new music generation tool within Google's Gemini AI. The presenter, having received early access and training from Google, aims to share insights and best practices for using Lyria to create high-quality music. The initial demonstration involves a basic prompt: "create music about the YouTube channel Ludo Salen." The presenter notes this is a "bad practice" as per Google's guidelines, predicting a mediocre result. Lyria generates a song titled "Le Reflet Salen." While the lyrics mention "béton" (concrete), which the presenter links to his hometown of Le Havre, the song doesn't specifically reference his YouTube channel. The presenter acknowledges the prompt was not optimized, leading to a "medium" and potentially "unusable" result outside of a casual setting. To illustrate a more effective workflow, the presenter first uses Gemini's image generation tool, Nano Banana, to create a promotional image for a fictional chocolate toothpaste called "Chocodent." This image generation process involves a detailed prompt to ensure high quality. The generated image, featuring a shining smile and the toothpaste tube, is then used as input for Lyria. The prompt for Lyria is again basic: "generate music for my ad." The intention is to create a catchy jingle. The presenter highlights that Lyria can process multiple images to create a music that accompanies a narrative, suggesting this could be useful for storytelling or children's entertainment. The resulting jingle, with the cover art featuring a toothpaste tube and the slogan "la pureté chocolat," is 43 seconds long. The presenter plays the song, which is in French. He notes that Lyria performs better in English, as indicated during his training. The jingle, though it incorporates "Chocoint Luxe" and "pureté chocolat," is described as "a bit medium" due to the lack of an optimized prompt. The video then delves into the core of the presenter's learning from the Google training: the "secret formula" or framework for crafting effective prompts for Lyria. This framework consists of five key components: 1. **Genre:** Specify the music genre and sub-genre (e.g., pop, rap, classical, film score, advertising music). Users can also combine genres or suggest specific eras (e.g., 80s, modern). A crucial tip here is how to evoke the style of specific artists without naming them directly due to copyright concerns. For instance, instead of saying "Taylor Swift," one should describe her characteristics, like "a female pop singer with a voice that dominates the charts in 2026." 2. **Tempo and Mood:** Describe the desired tempo (fast/slow, BPM) and the emotional tone (energy, melancholy, euphoria, dark, mysterious). If images are provided, Lyria will consider their ambiance. 3. **Instruments:** Specify up to two or three instruments. Google advises against listing too many instruments, as it can hinder Lyria's creativity. The AI can infer additional instruments if needed based on other prompt elements. 4. **Vocals:** Describe the vocal characteristics, such as gender (male/female), vocal quality (e.g., husky, smooth, smoker's voice), and emotional delivery (calm, shouting, reassuring, emotive). Specifying tessitura like soprano or baritone can also improve results. 5. **Lyrics:** Users can either provide their own lyrics or let Lyria generate them. The presenter suggests that providing lyrics separately can lead to more developed songs, as Lyria might generate shorter, less advanced music when it also has to create lyrics. The presenter reiterates that using simple, one-sentence prompts, as he did in his initial tests, results in generic, flat, and uninspired music with personality-less vocals. To achieve high-quality, unique results, an optimized prompt incorporating all five pillars is essential. Google presented several use cases during the training: * Creating music from one or more images, which can be a fun way to introduce AI to children by using their drawings. * Generating music for messages on SMS or WhatsApp, termed "sublimating daily routine" by Google. * Creating romantic music for a dinner by providing a restaurant receipt and a photo. * Generating music based on weather forecasts, like a rap song about Normandy's rainy weather. * Creating music based on personal characteristics or those of friends, using the example of someone who frequently forgets passwords. The presenter then introduces his own method for creating perfect prompts: using a custom "AI Box" or "Gem" within Gemini. He has trained his Gem with documents from his Lyria training and his notes. This Gem is designed to take a basic prompt and enhance it into an optimized, multi-part prompt for Lyria. He notes that prompts often work best in English, as many AI tools internally translate prompts to English. For a final test, the presenter uses his Gem to generate a song for the 2026 World Cup for the French national team, aiming for a fusion of pop and rap with a stadium anthem feel. The Gem produces an English prompt, which he then uses in Lyria. The output includes the song title, "La Troisème Étoile" (The Third Star), and the lyrics, which are played for the audience. The song is described as energetic, triumphant, and suitable for a stadium atmosphere. During the generation process, the presenter briefly promotes his private AI community, "QG IA," highlighting its benefits: a supportive community, help resources, training modules on AI tools, automation, and ROI from AI. He mentions that the community is hosted on the "School" platform, which bills in dollars. In conclusion, the presenter emphasizes the importance of structured, detailed prompts for Lyria to achieve high-quality, engaging music. He promises more AI tips, demos, and tutorials in future videos, aiming to empower viewers to leverage AI opportunities.