The Chinese AI lab DeepSeek recently released their new reasoning model R1, which is supposedly (a) better than the current best reasoning models (OpenAI’s o1- series), and (b) was trained on a GPU cluster a fraction the size of any of the big western AI labs.
DeepSeek uses a reinforcement learning approach, not a fine-tuning approach. There’s no need to generate a huge body of chain-of-thought data ahead of time, and there’s no need to run an expensive answer-checking model. Instead, the model generates its own chains-of-thought as it goes.
The secret behind their success? A bold move to train their models using FP8 (8-bit floating-point precision) instead of the standard FP32 (32-bit floating-point precision). … By using a clever system that applies high precision only when absolutely necessary, they achieved incredible efficiency without losing accuracy. … The impressive part? These multi-token predictions are about 85–90% accurate, meaning DeepSeek R1 can deliver high-quality answers at double the speed of its competitors.
a novel method for generating hyper-quality 4K textured mesh under only 30 seconds, providing 3D assets ready for commercial applications such as games, movies, and VR/AR.
Hunyuan3D 2.0, an advanced large-scale 3D synthesis system for generating high-resolution textured 3D assets. This system includes two foundation components: a large-scale shape generation model – Hunyuan3D-DiT, and a large-scale texture synthesis model – Hunyuan3D-Paint.
The shape generative model, built on a scalable flow-based diffusion transformer, aims to create geometry that properly aligns with a given condition image, laying a solid foundation for downstream applications. The texture synthesis model, benefiting from strong geometric and diffusion priors, produces high-resolution and vibrant texture maps for either generated or hand-crafted meshes. Furthermore, we build Hunyuan3D-Studio – a versatile, user-friendly production platform that simplifies the re-creation process of 3D assets.
It allows both professional and amateur users to manipulate or even animate their meshes efficiently. We systematically evaluate our models, showing that Hunyuan3D 2.0 outperforms previous state-of-the-art models, including the open-source models and closed-source models in geometry details, condition alignment, texture quality, and e.t.c.
Invoke is a powerful, secure, and easy-to-deploy generative AI platform for professional studios to create visual media. Train models on your intellectual property, control every aspect of the production process, and maintain complete ownership of your data, in perpetuity.
Stable Diffusion is a latent diffusion model that generates AI images from text. Instead of operating in the high-dimensional image space, it first compresses the image into the latent space.
Stable Diffusion belongs to a class of deep learning models called diffusion models. They are generative models, meaning they are designed to generate new data similar to what they have seen in training. In the case of Stable Diffusion, the data are images.
Why is it called the diffusion model? Because its math looks very much like diffusion in physics. Let’s go through the idea.
To reverse the diffusion, we need to know how much noise is added to an image. The answer is teaching a neural network model to predict the noise added. It is called the noise predictor in Stable Diffusion. It is a U-Net model.
After training, we have a noise predictor capable of estimating the noise added to an image.
Diffusion models like Google’s Imagen and Open AI’s DALL-E are in pixel space. They have used some tricks to make the model faster but still not enough.
Stable Diffusion is designed to solve the speed problem. Here’s how.
Stable Diffusion is a latent diffusion model. Instead of operating in the high-dimensional image space, it first compresses the image into the latent space. The latent space is 48 times smaller so it reaps the benefit of crunching a lot fewer numbers.
It is done using a technique called the variational autoencoder. Yes, that’s precisely what the VAE files are, but I will make it crystal clear later.
The Variational Autoencoder (VAE) neural network has two parts: (1) an encoder and (2) a decoder. The encoder compresses an image to a lower dimensional representation in the latent space. The decoder restores the image from the latent space.
You may wonder why the VAE can compress an image into a much smaller latent space without losing information. The reason is, unsurprisingly, natural images are not random. They have high regularity: A face follows a specific spatial relationship between the eyes, nose, cheek, and mouth. A dog has 4 legs and is a particular shape.
In other words, the high dimensionality of images is artifactual. Natural images can be readily compressed into the much smaller latent space without losing any information. This is called the manifold hypothesis in machine learning.
Where does the text prompt enter the picture?
This is where conditioning comes in. The purpose of conditioning is to steer the noise predictor so that the predicted noise will give us what we want after subtracting from the image.
The text prompt is not the only way a Stable Diffusion model can be conditioned. ControlNet conditions the noise predictor with detected outlines, human poses, etc, and achieves excellent controls over image generations.
This write-up won’t be complete without explaining Classifier-Free Guidance (CFG), a value AI artists tinker with every day. To understand what it is, we will need to first touch on its predecessor, classifier guidance…
The classifier guidance scale is a parameter for controlling how closely should the diffusion process follow the label.
Classifier-free guidance, in its authors’ terms, is a way to achieve “classifier guidance without a classifier”. They put the classifier part as conditioning of the noise predictor U-Net, achieving the so-called “classifier-free” (i.e., without a separate image classifier) guidance in image generation.
The SDXL model is the official upgrade to the v1 and v2 models. The model is released as open-source software. The total number of parameters of the SDXL model is 6.6 billion, compared with 0.98 billion for the v1.5 model.
The SDXL model is, in practice, two models. You run the base model, followed by the refiner model. The base model sets the global composition. The refiner model adds finer details.
This demonstrate large-scale text-to-video generation with a single neural function evaluation (1NFE) by using our proposed adversarial post-training technique. Our model generates 2 seconds of 1280×720 24fps videos in real-time
The IPAdapter are very powerful models for image-to-image conditioning. The subject or even just the style of the reference image(s) can be easily transferred to a generation. Think of it as a 1-image lora. They are an effective and lightweight adapter to achieve image prompt capability for the pre-trained text-to-image diffusion models. An IP-Adapter with only 22M parameters can achieve comparable or even better performance to a fine-tuned image prompt model.
Once the IP-Adapter is trained, it can be directly reusable on custom models fine-tuned from the same base model.
The IP-Adapter is fully compatible with existing controllable tools, e.g., ControlNet and T2I-Adapter.
SPAR3D is a fast single-image 3D reconstructor with intermediate point cloud generation, which allows for interactive user edits and achieves state-of-the-art performance.
MiniMax is thrilled to announce the release of the MiniMax-01 series, featuring two groundbreaking models:
MiniMax-Text-01: A foundational language model. MiniMax-VL-01: A visual multi-modal model.
Both models are now open-source, paving the way for innovation and accessibility in AI development!
🔑 Key Innovations
1. Lightning Attention Architecture: Combines 7/8 Lightning Attention with 1/8 Softmax Attention, delivering unparalleled performance.
2. Massive Scale with MoE (Mixture of Experts): 456B parameters with 32 experts and 45.9B activated parameters.
3. 4M-Token Context Window: Processes up to 4 million tokens, 20–32x the capacity of leading models, redefining what’s possible in long-context AI applications.
💡 Why MiniMax-01 Matters
1. Innovative Architecture for Top-Tier Performance
The MiniMax-01 series introduces the Lightning Attention mechanism, a bold alternative to traditional Transformer architectures, delivering unmatched efficiency and scalability.
2. 4M Ultra-Long Context: Ushering in the AI Agent Era
With the ability to handle 4 million tokens, MiniMax-01 is designed to lead the next wave of agent-based applications, where extended context handling and sustained memory are critical.
3. Unbeatable Cost-Effectiveness
Through proprietary architectural innovations and infrastructure optimization, we’re offering the most competitive pricing in the industry:
$0.2 per million input tokens
$1.1 per million output tokens
🌟 Experience the Future of AI Today
We believe MiniMax-01 is poised to transform AI applications across industries. Whether you’re building next-gen AI agents, tackling ultra-long context tasks, or exploring new frontiers in AI, MiniMax-01 is here to empower your vision.
DISCLAIMER – Links and images on this website may be protected by the respective owners’ copyright. All data submitted by users through this site shall be treated as freely available to share.
