ImageShield-MMCF — Multimodal Content Filter is a multimodal content-safety classifier built on top of Qwen3.5 and is now available on Hugging Face!
This is the preview initial version (v1.0) of the model, designed to classify visual content as Safe or Unsafe, with a particular focus on detecting Non-Consensual Intimate Imagery (NCII) and other potentially sensitive visual content.
The demo is implemented in the prithivMLmods/opencaption-4b-vl-sft Space, which serves as an active content-safety layer for computer vision tasks. It helps block NCII content generation and paves the way for more meaningful and responsible creativity.
If an agent can build the obvious demo, the obvious demo probably isn’t worth building anymore. For years, turning a research repo into something people could actually try was valuable by itself. That part is becoming automated — and that’s a good thing.
Which means the interesting work moves elsewhere: finding the weird use case, the right interaction, the unexpected model combination — or simply knowing which paper is worth anyone’s attention.
The demo used to be the product. Now it needs a point of view.
🚨 I've just published Sentence Transformers v6.0, introducing MultiVectorEncoder: ColBERT-style late interaction models are now a fourth model type, for training, inference, and interpretation, alongside the dense, sparse, and reranker models! Details:
Where a regular embedding model compresses a whole text into one vector, a multi-vector model keeps one vector per token and scores query against document with the MaxSim operator. That preserves token-level matching information that a single vector has to average away. It is also the state of the art for visual document retrieval, where a text query is matched against page images directly, charts and tables included, with no OCR step in between.
Any PyLate, Stanford ColBERT, or ColPali checkpoint loads straight into the same familiar API: model.encode_query(), model.encode_document(), and model.similarity() just work, whether the documents are texts or page images.
Does it help? LightOn trained LateOn (multi-vector) and DenseOn (dense) on the same data with the same 149M ModernBERT backbone, and the multi-vector model wins on 9 of the 13 NanoBEIR datasets: 0.6868 vs 0.6764 mean NDCG@10. The price is a bigger index, and the new HierarchicalTokenPooling module halves it at roughly no retrieval cost.
Antoine Chaffin, Raphaël Sourty, and I wrote a blog post walking through multi-vector models in practice: loading the various checkpoint formats, encoding and scoring, plugging them into a search stack, running them on page images, and keeping the index affordable. Check it out if you want to get started, or just point your Agent to the URL: https://huggingface.co/blog/multi-vector-encoder
**I benchmarked HF buckets against https access for Common Crawl.**
Took me a while to get round to do this but I benchmarked access to Common Crawl via https vs hf buckets. Both experiments were run at night in Europe. I do not think other hardware problems were impacting the speeds since CPU processing time of the non-download pipeline components were highly similar (within 2% identical) and below only the WarcReader speeds of datatrove are used.
Experiment: selected 5 disjoint samples of 64 files each (randomly from the latest crawl; 20,499 docs/file). Those five batches were then processed by 32 single-core tasks with 4GB/core (five batches to calculate CIs). Paired experiment between using https and hf bucket.
That is a difference of about 4x in streaming speed. You'll see that https is also more stable (smaller CI).
I also ran raw throughput tests to the endpoints to measure rate limiting (64MiB transfer at 8/32/128/256 concurrent readers) and rate limiting seems not an issue for either: at any of those parallel reader numbers, their respective speeds stay about the same.
Note that, given CC scale, this is still a small test. Rate limiting may become more obvious when processing a full crawl. I do not know whether the https endpoint vs HF bucket will shut you out earlier with which limits.
Made a demo for Text/Image-to-3D Video and Image-to-3D Video asset generation using TRELLIS.2. It is paired with Z-Image-Turbo to accelerate the input image preprocessing pipeline, streamlining the Image-to-3D workflow. The generated GLB (GL Transmission Format) files are converted into MP4 (MPEG-4) videos, making them easy to preview and share. Try it now on Hugging Face Spaces.🤗
⏱️ Built a small Space for Visual Chronometer / Pulse of Motion.
Upload a video and estimate its Physical FPS: the frame rate implied by visual motion, independent of metadata. Useful to inspect “chronometric hallucination” in generated videos: clips that look smooth, but move with the wrong physical time scale.
A few weeks ago, @victor opened the door: coding agents can now ship Hugging Face Spaces autonomously.
I pulled on that thread.
As someone who builds and ships Gradio demos regularly, I didn’t just want to reproduce the loop. I wanted to see what happens when that loop is plugged into the whole Hugging Face stack.
The interesting part is not only that an agent can ship a Space.
It’s what happens when Space generation becomes a first-class Hugging Face workflow.
Wan2.2-I2V-Fast with highly upscaled sequential frame sampling is now available as a Spaces demo, built using Wan2.2-I2V and FLUX.2-Klein. Try the demo using the links below.👇
New blog post! An introduction to a little-known but highly effective model reduction method: 𝗧𝗿𝗶𝗺𝗺𝗶𝗻𝗴✂️ We show how to reduce model size (we went up to 87.24% reduction) while preserving its performance.
We applied this technique to 16 different model families across several modalities to illustrate that it works on any architecture (as long as the embedding layer is the last one of the model) and on any modality involving text. From these 16 families, we generated over 𝟱,𝟱𝟬𝟬 𝗺𝗼𝗻𝗼𝗹𝗶𝗻𝗴𝘂𝗮𝗹 𝗺𝗼𝗱𝗲𝗹𝘀 𝗶𝗻 𝟭𝟮𝟰 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲𝘀 🌍
Key takeaways from our experiments: 1️⃣ Trimming does not require a GPU. Our models were obtained on a CPU. 2️⃣ This method scales up to at least 4B parameters (we did not test beyond that). 3️⃣ Trimmed model is smaller than the original while preserving its performance. If you observe a slight performance drop, just fine-tuned to recover or even surpass the original performance. 4️⃣ For an equivalent compute budget, it is better to trim then fine-tune rather than fine-tuning the original model. Since the model is smaller, you can run more epochs/show more data and get in fine a better model than the original. 5️⃣ Trimming is a competitive alternative to distillation and quantization. E.g. we obtained our alternative to DistilBERT in 9 minutes on CPU vs. 90 hours of GPU for the latter. 6️⃣ Trimming could generate reasoning traces in the language of the trimmed model. This could be an alternative to generating traces in English and then translating them into the desired language.
And many other things (such as how much data are needed, the impact of the database used, the order in which it should be done, etc.) are available in the blogpost!
PiD — Pixel Diffusion Decoder Image Edit Upscale and Image Generation Upscale, an all-in-one demo, is now live on Spaces! Great improvements in realism-based image generation and editing are powered by FLUX.2-Klein, while image generation is paired with Z-Image, and upscaling is enabled by default!
I've made 8 Spaces in the Qwen-Image-Edit series, and out of them, 5 Spaces reached “Space of the Week”! A few Spaces are still topping the list even after many months.
Cumulatively, the series has crossed 8.2 million+ ZeroGPU runs and nearly 4 million visitors overall.
I built a little demo where you give three models (Apertus, Llama, Qwen3) the same prompt and in the end you have to guess which is which just based on their answers.
🤗 Announcing the Ettin Reranker family: six new state-of-the-art CrossEncoder rerankers for search from 17M to 1B parameters, plus the full training data and the ~150-line recipe. Built on the Ettin ModernBERT encoders, Apache 2.0. Details:
All six were trained with the same single-stage pointwise MSE distillation recipe, with mixedbread-ai/mxbai-rerank-large-v2 (1.54B) as the teacher. Only the learning rate and per-device batch size change between sizes. The 1B student matches the teacher within 0.0001 NDCG@10 on MTEB(eng, v2) Retrieval, the 150M is the strongest reranker I tested in the under-600M range, and the 17M beats the 33M ms-marco-MiniLM-L12-v2 by +0.051 NDCG@10 at roughly half the parameter count.
Speed matters as much as quality for a reranker, since it determines whether the model fits the latency budget between retrieval and showing results. Our 17M is the fastest reranker in the whole comparison at 7517 pairs/sec on an H100. Our 150M runs 2.3x faster than the two other 150M ModernBERT-base rerankers (gte-reranker-modernbert-base and granite-embedding-reranker-english-r2) because the modular Transformer module propagates unpadded inputs through every layer rather than just the FA2 attention kernel. And our 1B is 2.4x faster than its 1.5B teacher while matching it on quality.
I bootstrapped the training recipe with the new train-sentence-transformers Agent Skill shipped in Sentence Transformers v5.5.0. Install it with hf skills add train-sentence-transformers --claude and ask Claude Code (or Codex / Cursor / Gemini CLI) to fine-tune a SentenceTransformer, CrossEncoder, or SparseEncoder model on your data.
I wrote a blog post walking through usage, results across six embedder pairings, the speed story, and the complete training script. Check it out, or just point your Agent to the URL:
A live community radio for AI-generated songs, powered by tracks created with ACE-Step.
You can tune in, discover community-made songs in many languages, vote on what sounds good, and mark your real favorites as Bangers.
The more people listen, vote, and create, the better the station gets.
Under the hood, it connects a few Hugging Face pieces together:
Spaces for the live app, HF buckets for community tracks, OAuth for signed-in listeners, server-side streaming with ffmpeg, hourly playlist refreshes, moderation, jingles, and community feedback loops.
It’s not just a playlist.
It’s a shared taste experiment: new songs get a shot every hour, and the community helps decide what deserves another spin.
Come listen. Find weird gems. Support the Bangers. Shape the radio.
Great technical guide by Nico Martin on the Hugging Face blog, showing how to use Transformers.js inside a Chrome extension and run ONNX models from the Hub locally with WebGPU inside a Manifest V3 extension.
The interesting part: this is not just a chatbot in a side panel.
The article walks through the architecture behind a browser agent that can read open tabs, query webpages, search history, and highlight elements directly on the page — with models downloaded from the Hugging Face Hub, cached under the extension origin, and executed locally instead of being called through a remote API for every prompt.
A strong blueprint for building local-first web copilots, reading assistants, and AI-powered browsing workflows.
🤖 I've just published Sentence Transformers v5.5.0, headlined by a new train-sentence-transformers Agent Skill that lets your AI coding agent (Claude Code, Codex, Cursor, Gemini CLI, ...) train and finetune embedding, reranker, and sparse encoder models for you. Plus training losses & fixes. Details:
The skill bundles curated guidance for the whole training workflow across all three model types: base model selection, loss and evaluator choice, hard-negative mining, distillation, LoRA, Matryoshka, multilingual training, static embeddings, etc. It also ships production-ready training template scripts the agent can adapt. Install it with hf skills add train-sentence-transformers, then just describe what you want, e.g. "finetune a reranker on my (question, answer) pairs, mine hard negatives, and push it to the Hub".
On the loss side: EmbedDistillLoss is a new embedding-level distillation loss for SentenceTransformer. Instead of distilling teacher scores like MarginMSELoss, it aligns the student's embeddings directly with pre-computed teacher embeddings, wtih an optional learnable projection for when the student and teacher dimensions differ. Second, ADRMSELoss is a new listwise learning-to-rank loss for CrossEncoder from the Rank-DistilLLM paper, aimed at the LLM-distillation reranking setting.
encode() and predict() also gained a per-call processing_kwargs override, so you can change processor settings like max_length, a vision-language model's image resolution, or a video's fps, for a single call without rebuilding the model.
The Agent Skill is the part of this release I'm most keen for people to try. Curious to hear how it works for you. I've been using it myself a lot to quickly set up some training runs that immediately use a bunch of best practices.
