Anticipating a Surge in Deepfakes: The Impact of AI Image Generation Advancements (Farewell, LoRA?)

Anticipating a Surge in Deepfakes: The Impact of AI Image Generation Advancements (Farewell, LoRA?)

Subscribe to our daily and weekly newsletters for the latest updates and exclusive content on top-tier AI coverage.

A new AI image generation method called InstantID can quickly identify a person and create new images based on just one reference photo. This is highlighted in a recent paper by the InstantX team in Beijing.

However, Reuven Cohen, an enterprise AI consultant for Fortune 500 companies, has pointed out a significant drawback: InstantID could lead to an influx of deepfake audio, images, and video tools, especially with the 2024 election approaching.

Cohen notes, “Using tools like InstantID for deepfakes is concerning because they are easy to use and produce consistent results without the need for training or fine-tuning. The ability of InstantID to generate identity-preserving content means highly realistic and convincing deepfakes can be made with minimal resources.”

InstantID outperforms LoRA in generating recognizable AI images. According to Cohen, InstantID is superior to LoRA, which involves small, fine-tuned models built on limited parameters like specific characters or styles. While LoRA has led to various creations shared on controversial platforms like Civitai — including AI-generated fan fiction, anime characters, photorealism, fashion, and notably, pornographic content and deepfakes — InstantID surpasses it in capability.

Cohen shared his thoughts on LinkedIn, claiming InstantID as the new front-runner over LoRA, referring to it as “deep fakes on steroids.”

The InstantX team’s paper titled “InstantID: Zero-shot Identity-Preserving Generation in Seconds” mentions that techniques like LoRA are limited by their high storage needs, lengthy fine-tuning, and multiple reference images. Current ID embedding methods also have their challenges, but InstantID provides a ‘plug and play module’ that adeptly personalizes images in various styles using just a single facial image while maintaining high fidelity.

Cohen explained that InstantID enables zero-shot identity-preserving generation, unlike LoRA and QLoRA. QLoRA improves upon LoRA by simplifying the model’s data, thus reducing the resources required for fine-tuning. Cohen stated, “LoRA and QLoRA are about fine-tuning part of the model parameters or applying quantization for efficiency, but InstantID focuses on fast and efficient identity-preserving output.”

Creating AI deepfakes is now simpler than ever. InstantID’s main function is not about fine-tuning models but maintaining identity consistency in generated content, Cohen added. For example, “Donald Trump always looks like Donald Trump.”

Cohen warns that it has never been easier to quickly create deepfakes, stating, “It’s just one click to deploy this on platforms like Hugging Face or replicate.”

Stay informed by subscribing to our daily newsletter and get the latest news straight to your inbox.