AI tools have the capability to produce images of bespoke digital identities, however, this requires detailed adjustment of Low-Rank Adaptation (LoRA) to achieve quality outcomes. A new entrant to this space, InstantID, bypasses this complication with a zero-shot plug-in that allows generative AI models to produce consistent images using just a singular reference face image.
Conventionally, quality image generation of a particular individual via a generative model requires the application of a LoRA. The technique, which allows the adaptation of image generation models without fully retraining them, takes up considerable resources in terms of time, reference images, and processing capacity. It is here that InstantID comes into play, which could potentially lead to discontinuation of LoRAs for many applications.
Developed by the InstantX Team, InstantID is a zero-shot model that necessitates no training or tuning. Utilizing just a single face for reference, it is capable of guiding a text-to-image Stable Diffusion model, such as SD1.5 or SDXL, to produce additional images of the person. The key lies in an IdentityNet component that emphasizes specific facial elements present in the reference image over other elements.
One of the major advantages of InstantID is its capability for consistent character generation. This is particularly useful in scenarios where one needs to generate images of a character for a game or graphic novel, where it is often challenging to get an AI image generator to maintain uniformity in the character’s facial features. InstantID strikes a balance between stylistic and other generative functions while preserving high-fidelity facial features.
On the flip side, InstantID also introduces significant risks. For example, on controversial platforms like Civitai, LoRAs are widely used to create AI-generated adult content. InstantID is likely to further enable the creation of such content as it removes the necessity of a LoRA or access to substantial cloud computing power to generate a realistic image of a specific person. All it needs is one photograph.
Moreover, there have been unintended consequences, evidenced by the unfortunate use of Taylor Swift’s images as examples in some of the initial applications. The proliferation of inappropriate Taylor Swift images across the internet is a sign of potential future misuse. The creators of InstantID have acknowledged this issue, noting that it could enable “the potential creation of offensive or culturally inappropriate imagery”.
In a nutshell, while InstantID opens up exciting possibilities in AI image generation from a single face image, it also raises significant concerns about potential misuse.
