Image Tool Pikaso Now Uses Multiple References

- Freepik's "Custom Characters" feature is powered by LoRA (Low-Rank Adaptation) technology, which fine-tunes the AI model on a user-provided dataset. This allows the AI to learn a specific character's features from a set of 12 to 24 uploaded images, enabling it to generate that character in various poses and scenes. - This method of training a custom model on a character is part of a broader trend in generative AI to give users more granular control and solve the critical issue of character consistency across multiple generated images. This addresses a common frustration for creators using AI for narrative projects like storyboards or comics. - The trained "Custom Character" can be integrated into a larger creative workflow within Freepik's ecosystem. For instance, a character generated can then be used as a base for Freepik's AI Video Generator, creating a pipeline from a static image to animated content. - The approach of using multiple image references is becoming a key differentiator among AI image tools. While Pikaso uses a training-based LoRA model, other platforms like LTX Studio use models such as FLUX Kontext, which allow for the combination of up to two reference images with text prompts to guide elements like pose and style in real-time. - This level of control, where users train the AI on their own curated datasets or combine multiple specific references, is central to the ongoing debate about authorship in AI-assisted art. It shifts the user's role from simply writing a text prompt to actively curating the visual data that guides the AI, strengthening the argument for human authorship. - For developers and builders, the integration of such features is often handled via APIs that allow for the programmatic creation and management of these custom character models, sometimes referred to as LoRAs. This opens the door for building specialized creative tools on top of foundational image generation models.