fal’s Post

These images, which quotes from one of my papers, are created 'on the fly', rather than having to create them all manually, which would be too time consuming. #GXKSoundbite @micro_mart @pexels https://buff.ly/3AuuyPu

for the real one go to https://casamamaiapasa.ro/ *image generated using Dalle 3

🤔 Do we need separate models for image editing and conditional generation? We introduce🌟EditAR (https://lnkd.in/g5n33KXC), a unified autoregressive model for diverse tasks, e.g., image editing, depth-to-image, edge-to-image, segmentation-to-image.

Autoregressive models are picking up for image generation, but how about image editing? Given an image and a language description, we train one Autoregressive Transformer to do ANY EDITING.

I added LLM support to Cremage (https://lnkd.in/g5Mmkssw) so that you can discuss images generated using Cremage or generated externally with an LLM running locally ("llava-hf/llava-v1.6-mistral-7b-hf"). Check out the video below to see how it works. The video also shows my pencil drawing of a puppy to demo the LLM can handle non-generative-AI images.

How often were you asked generic questions, such as "Enhance this video" and "Find me everything that is interesting in this hard drive"? It happened to me many times, and the first principle is dedicated to this issue. #videoevidenceillustrated PRINCIPLE 1. HAVE A SPECIFIC QUESTION AND PURPOSE FOR THE MEDIA Analyzing an image or video is just a means to an end. It is necessary to know what we are looking for. It doesn’t make sense to forensically “enhance” an image without a clear purpose. Most of the time, processing an image will deliver a less pleasant result globally but provide more clarity on the parts of the image where the relevant information is located. For example, if we enhance a license plate, we often lose quality in other parts of the image. Generally, the two most common requests during video investigations are the identification (typically of a person or a vehicle) and the reconstruction and understanding of a sequence of events. Following this principle will drive the analysis in the right direction, waste less time, and have a better outcome. How can we answer a question if we don’t know what we are looking for? Check out the full video evidence principles document here: https://lnkd.in/dcup53nn

Most #scams are virtually undetectable by the human eye. 👁️ Scamnetic’s Scan & Score gives you the information you need to communicate with confidence. Our patent-pending process scans emails, texts, images, URL, and more to deliver clear, detailed results and recommendations to stay safe. Find out more: https://lnkd.in/dZ9FwAZ5

We have just created a new guide to Transition Models. This graphic illustrates some of the key dimensions common to different keys of transitions within complex systems. You can find the full guide here: https://lnkd.in/edYB8CJW

Can We Generate Images with CoT? Let's Verify and Reinforce Image Generation Step by Step Guo et al.: https://lnkd.in/epFpeWPQ #ArtificialIntelligence #DeepLearning #MachineLearning

(1) How can we combine generative models, such as those for backgrounds, with models of faces? (2) How can we achieve continuous rig-like controls for text-to-image generation?  (3) How can we enhance the performance of zero-shot image editors? Explore our contrastive guidance paper, a unified approach to classifier-free guidance methods that can address many of these challenges. https://lnkd.in/gYv7ybpF