A new AI image generation technique called InstantID allows for rapid identification and creation of images based on a single reference image, according to a recent paper by the InstantX team in Beijing.
Reuven Cohen, an enterprise AI consultant for Fortune 500 companies, refers to InstantID as the “new state-of-the-art” in AI image generation. However, he warns that this technology could lead to a surge in deepfake content—audio, images, and videos—especially with the 2024 elections approaching.
Cohen commented, “The use of tools like InstantID for deepfakes raises significant concerns due to the ease of creation and the consistency of output, requiring no training or fine-tuning.” He highlighted that InstantID can produce highly realistic deepfakes with minimal computational resources: “It can efficiently generate identity-preserving content with little CPU and no GPU power needed.”
InstantID vs. LoRA: A Major Advancement
Cohen explains that InstantID outperforms LoRA, which involves small, fine-tuned models trained on limited parameters such as specific characters or artistic styles. While LoRA has enabled a wide array of creations, from AI-generated fan fiction to photorealism, it is controversially best known for producing pornography and deepfakes.
In a LinkedIn post, Cohen remarked, “So long, LoRA,” noting that InstantID represents “deepfakes on steroids.”
The InstantX team’s paper, titled InstantID: Zero-shot Identity-Preserving Generation in Seconds, states that existing methods like LoRA face limitations due to high storage needs, extensive fine-tuning, and the requirement of multiple reference images. In contrast, InstantID provides a ‘plug and play module’ that efficiently personalizes images in various styles using just one facial image, all while ensuring high fidelity.
Cohen explains that InstantID is designed for zero-shot identity-preserving generation, which is fundamentally different from existing techniques like QLoRA that simplify model data to decrease resource requirements for fine-tuning. While QLoRA was previously the cutting-edge method, he emphasizes that InstantID’s focus is on swiftly generating outputs that retain the identity characteristics of the input data.
Simplifying Deepfake Creation
InstantID’s primary function is to maintain the identity of individuals in generated content. “Think about consistency—like how Donald Trump always looks like Donald Trump,” he noted. He cautioned that creating deepfakes has never been easier: “With just one click, you can deploy this on Hugging Face or replicate it.”
As technology evolves, the implications of accessible deepfake tools like InstantID are vast, raising important questions about authenticity and the future of digital content.