AI ML DL

FaceXFormer
AI ML DL

A Unified System for Facial Analysis

A new offering from John Hopkins University aims to integrate essential facial analysis tasks, such as semantic segmentation, landmark attribution, and age, race and gender classification, into a single framework, powered by a novel Transformer-based token system.

Read More »
Animate Your Motion
AI ML DL

The Continuing Struggle to Create Significant Motion in Text-To-Video Output

Text-to-video systems such as Sora and Runway’s offerings wow the internet, but actually feature very limited real movement. Only a small number of current publications deal with the problem, such as the recent Boximator project, which uses bounding boxes as movement targets and forces more obvious movement shifts. Now, a new system offers a multimodal approach to producing ‘abrupt’ movements in generated video.

Read More »
SuDe
AI ML DL

Better Stable Diffusion Deepfakes With an Object-Oriented Programming Approach

New research out of China finds that Stable Diffusion is far more capable of accurate customization if a small but essential change is made to the way that the information is trained into the adjunct model. With wider adoption, the new method could offer a revolution in text-to-image fidelity, by addressing a fundamental flaw in the original architecture of customization methods.

Read More »
LayerDiffuse
AI ML DL

Getting Real Transparency Into Stable Diffusion

A new system from Stanford University has retrofitted Stable Diffusion so that it is almost natively capable of producing images with alpha channels, offering the potential for more controllable output in neural workflows for visual effects houses. Titled LayerDiffuse, the new approach also scores competitively against professional and quite expensive stock photo output.

Read More »
LatentSwap
AI ML DL

Face-Swapping Directly in the Latent Space

A new collaboration between various institutions and companies in Korea and MIT offers a superior method of deepfaking, using direct manipulation of the latent space, with no need to provide source material, or to fine-tune existing models. The method uses several prior frameworks, including StyleGAN2, and operates with only 87 million parameters, compared to the average 200 million parameters of former methods.

Read More »
EMO
AI ML DL

Plausible Stable Diffusion Video From a Single Image

The new EMO system, from the Alibaba Group, has made a notable impression on the internet, with its ability to create highly realistic talking and singing videos from just a single source image, using Stable Diffusion and the latest and most popular adjunct technologies. But is it good for anything more than the viral ‘wow’ factor..?

Read More »
One2Avatar examples
AI ML DL

Better Neural Avatars From Just Five Face Images

Many neural avatar systems of the last 18 months require extensive training data, or even full videoclips. Others are performant, but have exorbitant training demands. However, a new system from Google and the University of Minnesota is proposing a photorealistic deepfake head system that’s trained on only five images – and can work quite well from just one image; and the new system of pretraining that the framework uses throws some of the conventions regarding hyperscale training datasets into question.

Read More »
AI ML DL

The Challenge of Preventing ‘Identity Bleed’ in Face Swaps

KAIST AI has developed a new method of disentangling identity characteristics in a face-swap from secondary characteristics such as lighting, skin texture – and the original structure of the face to be ‘overwritten’ by the new identity. If such techniques can be perfected, facial replacement could be freed from having the original identity ‘bleeding through’ into the superimposed identity.

Read More »
Fake faces broken down into the spectral ranges that may reveal them
AI ML DL

Combating Stable Diffusion Face Forgery Through Frequency Analysis

For over six years, deepfake detection methods have sought to find a criteria for detection that can survive the evolution of generative systems. A new paper from Switzerland and France offers such a method, by examining the way that frequency spectra differ among real and generated images.

Read More »