¡Google acaba de Destrozar los Influencers! Crea personajes consistentes GRATIS
The video demonstrates how to use Google Flow to create consistent AI characters for free, covering four key elements: consistency, variations, movement, and style transfer. The presenter walks through creating characters with detailed prompts, animating them with voice assignments, and transferring movements using Gemini's OVNI model. Free alternatives like OneVideo are also mentioned for users who exhaust Flow's credit limits.
Summary
The video opens by highlighting the growing trend of AI-generated virtual influencers, noting that Spanish creators and major brands like IKEA are already leveraging non-existent AI characters for commercial purposes. The presenter frames Google Flow as a newly updated platform that can facilitate this creation process largely for free, organizing the tutorial around four pillars: consistency, variations, movement, and style transfer.
In the first section on consistency, the presenter introduces Google Flow's 'Characters' feature, which saves AI-generated characters for reuse across projects. A nine-pillar prompt framework is outlined for creating realistic characters, covering shot type, technical realism, unique identifiers, facial structure, skin texture, lighting, clothing, composition, and emotional tone. A custom Gemini Gem template is offered to automate this process. The presenter also demonstrates how uploading a personal photo allows users to create a realistic avatar of themselves, with a three-perspective (front, side, back) image system ensuring full-body consistency.
The second section covers variations, showing how the '@' mention system in Flow lets users apply saved characters to new scenes — such as teaching a class of robot students or posing with a perfume product. The presenter demonstrates placing multiple characters in a single scene, as well as using Flow's new Tools suite for image editing, angle changes, brightness/contrast adjustments, text overlays, in-painting, and outpainting for format resizing.
The third section addresses movement and video generation. Flow's video feature uses credits (50 renewed daily on free plans), and the presenter demonstrates assigning a consistent AI voice to a character so all generated videos use the same vocal identity. Two videos are generated to verify voice consistency across different scenes. Techniques for improving video quality are discussed, including using the 'Fast' model and setting start/end frames for more controlled output.
The fourth section explores style and movement transfer using Gemini's OVNI model, which accepts multiple video inputs and blends them. The presenter shows replacing themselves in a selfie video with a Mario Bros. scene, changing their setting to a TV studio, and transferring their own gestures onto the science character. OneVideo is introduced as a fully free alternative platform with avatar animation and character-switching capabilities similar to Gemini OVNI, concluding with a philosophical note that in the age of AI characters, the message matters more than the face delivering it.
Key Insights
- The presenter argues that providing a highly detailed nine-pillar prompt — covering shot type, realism level, unique identifiers, facial structure, skin texture, lighting, clothing, composition, and emotional tone — is what separates realistic AI character images from obviously AI-generated ones.
- Google Flow's character creation and image variation features are described as completely free and unlimited, while video generation uses a credit system that refreshes at 50 credits per day even on free plans.
- The presenter demonstrates that assigning a consistent AI-generated voice to a saved character in Flow ensures vocal identity is preserved across all future videos featuring that character, enabling cross-scene voice consistency.
- Gemini's OVNI model is shown to accept multiple video inputs simultaneously and can transfer movement from one person's gestures onto a previously created AI character, effectively replacing the performer's likeness while preserving motion.
- The presenter notes that setting a VPN to the United States or switching prompts to English helps bypass policy compliance rejections when using the OVNI model in Google Flow.
Topics
Full transcript available for MurmurCast members
Sign Up to Access