News : The ternary-bonsai model with hermes tui interface on huggingface.co website.

Today, we’re expanding the Bonsai 2 family with three new ternary models based on Qwen3.5: Ternary Bonsai 2 2B, 4B, and 9B. Joining Bonsai 2 27B, they bring stronger capabilities to smaller deployment footprints and give developers more choices for building AI applications within local devices.

Building on first-generation Bonsai 1.7B, 4B, and 8B, the new models combine improved reasoning, visual understanding, and support for tool-driven workflows. The family now spans four sizes, allowing developers to balance task complexity, response time, and memory requirements.

The new models use a ternary representation that encodes weights using three values: {−1, 0, +1}. The 2B, 4B and 9B models have footprints of 0.43GB, 0.95GB and 2GB, respectively, more than 9× smaller than their full-precision counterparts.

About these models you can read on the official website.

See this space on huggingface.co how can use it with hermes tui interface.

News : DeepSite is moving to Hugging Face Chat.

We are migrating DeepSite into Hugging Face Chat to deliver a better long-term experience. It will be actively maintained, regularly improved, and always ready with the latest models.

New DeepSeek V4.1 Flash

DeepSeek's natively multimodal 552B MoE, with frontier-level agentic coding and a 1M-token context window.

See the old website with redirect the new one.

News : PointDiT: Pixel-Space Diffusion for Monocular Geometry Estimation

State-of-the-art single-image 3D reconstruction methods often rely on complex hybrid architectures and loss functions, or compress geometry into latent spaces in order to leverage pre-trained latent diffusion models. In this work, we show that such architectural overhead and intricate loss formulations are unnecessary. We introduce a minimalist pixel-space Diffusion Transformer, built on a plain ViT, that operates directly on raw 3D point map patches and is conditioned on image tokens from a pre-trained DINOv3. Unlike existing latent diffusion approaches, we train our diffusion backbone entirely from scratch, eliminating the need for point map tokenizers. Despite its simplicity, our approach surpasses complex latent-based diffusion models while remaining significantly simpler than hybrid alternatives. Notably, it produces sharper geometric structure and is more robust in highly ambiguous regions, such as transparent objects.

See the official huggingface.co website.

News : Wan2.2-Animate animation and replacement.

Wan2.2-Animate: Unified Character Animation and Replacement with Holistic Replication.

You can test on the demo - huggingface.co.

Tools : new Qwen-Image-Edit-2511-Multiple-Angles-LoRA on the HuggingFace webpage.

Today, I see this new Qwen-Image-Edit-2511-Multiple-Angles-LoRA implementation on the HuggingFace.co website. Also, I used the demo reel to test with one web image with one comic - Iron Man.

See this huggingface.co page

News : Image edit with Qwen-Edit-2509-Multiple-angles LoRA and Lightning LoRA.

Qwen Image Edit 2509 for Camera Control ✨ Using dx8152's Qwen-Edit-2509-Multiple-angles LoRA and Phr00t/Qwen-Image-Edit-Rapid-AIO for 4-step inference 💨

... and Fast 4 steps Wan 2.2 I2V (14B) with Lightning LoRA

I tested first tool with default example on this website page, you can find another example on this website page.

News : OpenMed - open-source AI models.

OpenMed is new open-source AI models and datasets made to help advance healthcare and medical research. This provides tools, resources, and community support for anyone interested in using AI for tasks like medical imaging, diagnostics, and data analysis.

You can find it on the Hugging Face website.

News : Hunyuan3D-2mini-Turbo: Fast Strong 0.6B Image to Shape Generator - online tool.

See this tool on the huggingface.co - official website.

News : Testing GeometryCrafter: Consistent Geometry Estimation for Open-world Videos with Diffusion Priors

The paper arXiv:2504.01016 introduces GeometryCrafter, a framework designed to estimate temporally consistent, high-quality point maps from open-world videos. These point maps are crucial for applications like 3D/4D reconstruction, depth-based video editing, and camera parameter estimation.

I tested this week on the huggingface.co GeometryCrafter project and works with a short video because I don't have a Pro account ...

The GeometryCrafter can generates .npz files. These are NumPy compressed archive files that store the predicted point maps corresponding to each frame of the input video. Each .npz file encapsulates the 3D spatial information derived from the video frames, facilitating further processing or visualization.

I tested a python script for .npz with Blender 3D software and works , not finished yet!

News : huggingface providers feature ...

... I tested few days ago these providders in the huggingface website, not all working well.

The fal provider work with python module in Windows 10 with setx and FAL_KEY, I can tell for what reason ... but is a good idea to have artificial intelligence providers into the huggingface website.

News : FacePoke by jbilcke-hf at huggingface.co .

You can test with this online demo by jbilcke-hf ...

Huggingface : Try Off Anyone

NOTE : I worked a little with artificial intelligence, then I studied some of the theories and various A.P.I. that I used at the first steps level. It is not a very complex mathematics, the implementation ideas derive from this mathematics and/or from the similar way of processing.

Today I decided to open a series of articles related to huggingface.co website because it has a lot of information and examples implemented with artificial intelligence.

And because it is easier to go through concrete and already processed information, here is an example for today: extracting an article of clothing from a photo.

You can test the demo online at the website - huggingface.co.

LivePortrait : Add mimics and lip sync to your static portrait driven by a video ...

You can tst online on the huggingface.co - KwaiVGI - LivePortrait

The source code can be found on this GitHub repo.

A demo real video from theAIsearch:

News : DreaMoving AI customize your video.

... DreaMoving, a diffusion-based controllable video generation framework to produce high-quality customized human videos.

You can read more about this artificial intelligence on this official webpage.

If you want to test this feature then you can use the huggingface website.

News : Depth Prediction - Image to 3D - 360 Panorama to 3D with ZoeDepth.

ZoeDepth means Zero-shot Transfer by Combining Relative and Metric Depth is not new in the area of processing images and artificial intelligence but this online tool shows how this works.

Open LLM Leaderboard for artificial intelligence.

The Open LLM Leaderboard tracks, ranks, and evaluates large language models and chatbots. It evaluates models based on benchmarks from the Eleuther AI Language Model Evaluation Harness, covering science questions, commonsense inference, multitask accuracy, and truthfulness in generating answers.

see all apps on the official website.