News
- https://youtu.be/b9cFjfcwoGw
Commentary
Virtual influencers, also known as "virtual influencers" or "vfluencers," are digital characters designed to interact with audiences on social media platforms such as Instagram, TikTok, and YouTube. These digital avatars are created using AI algorithms, 3D modeling, and rendering techniques to closely resemble real human beings. From their appearance to their personality and lifestyle, every aspect of these influencers is carefully crafted to attract and engage online users.
One of the advantages of this type of digital avatars is their ability to operate indefinitely, 24 hours a day, without requiring rest. Moreover, they can be controlled and directed to precisely align with marketing strategies. Brands can leverage this technology to create personalized and segmented advertising campaigns, taking advantage of AI’s capacity to analyze data and understand audience preferences. However, the emergence of this technology also raises significant ethical and social concerns. To what extent are these digital avatars contributing to authenticity and transparency in the online world? Are we not creating a fictional environment that has nothing to do with reality? What impact will they have on public perception of identity on social media? Furthermore, complete corporate control over these figures could generate issues related to privacy and manipulation.
In this interesting video by Adrián Sánchez and Ramón, the business of AI-generated influencers is reviewed, and even more intriguingly, how they are created using applications such as Fooocus, which are freely available for download and testing on GitHub.
Fooocus is an image generation application based on the Stable Diffusion XL architecture, designed with a radically different approach compared to other similar tools. While platforms such as Automatic1111 or ComfyUI offer granular control over every parameter of the generation process, Fooocus prioritizes simplicity: users need only focus on the prompt and the resulting image, without requiring complex manual adjustments. The tool is offline, open-source, and free, with installation reduced to less than three clicks from download to the first generated image.
The creation of virtual influencers using Fooocus leverages several of its standout features. The application includes a GPT-2-based prompt processing engine that expands user descriptions to enhance image quality, a methodology similar to that employed by commercial services such as Midjourney or LeonardoAI. Additionally, it incorporates specialized algorithms for inpainting and outpainting that enable modification of specific image regions or extension of their borders—essential functionalities for refining the appearance of a digital character. The tool also offers face-swapping capabilities via InsightFace, facilitating facial consistency across different generations.
From a technical perspective, Fooocus implements several innovations that distinguish it from other Stable Diffusion implementations. These include a native refiner exchange system within a single k-sampler, enabling a more coherent transition between low- and high-resolution generation phases; negative ADM guidance that compensates for the lack of contrast in the model’s higher-resolution layer; and an adjusted variant of self-attention guidance that avoids the overly smooth or plastic appearance sometimes produced by SDXL. These improvements, though invisible to the user, are decisive in achieving an appropriate level of realism in the representation of faces and human figures.
Creating a virtual influencer with Fooocus typically involves multiple stages. First, base images are generated using prompts that describe the desired appearance: facial features, hairstyle, clothing, and artistic style. Next, details are refined through inpainting to correct imperfections or add specific elements. Subsequently, the face swap function is employed to maintain facial consistency across different images and scenarios. Finally, variations of the character in various poses, environments, and situations are generated, thereby creating a cohesive set of visual content that simulates a consistent presence on social media.
The phenomenon of virtual influencers, enabled by accessible tools such as Fooocus, represents an evolution of trends already observed in previous articles of this series. Just as customized GPTs allowed for the specialization of language models for specific tasks, Fooocus democratizes the creation of hyperrealistic visual content, granting non-technical users capabilities that previously required dedicated 3D design and rendering teams. However, this accessibility amplifies the tensions already identified between technological innovation and ethical responsibility.
The ease with which synthetic human characters can be generated raises questions about transparency in commercial communication. If a follower interacts with a virtual influencer without being aware of its artificial nature, is their right to accurate information being violated? Social platforms have begun to require labeling of AI-generated content, but the effective enforcement of these regulations remains inconsistent. Moreover, the ability to create characters that simulate specific identities—characterized by particular ethnic traits, ages, or cultural contexts—demands reflection on representation and the potential for commercial entities to appropriate identities.
The existence of this type of tools, available free of charge, also raises a question of sustainability. The project is currently in a state of limited long-term support, with updates restricted to bug fixes and no immediate plans to migrate to new architectures such as Flux. This status reflects a common dynamic in the Open Source ecosystem: initial development driven by enthusiasts reaches a level of maturity that enables stable use, but further evolution depends on the availability of resources and community interest.