#多模态
200
2026-09-09
Show HN: Sparrow-2 – Noise cancellation isn't designed for conversational AI
S
Hey there, I’m Brian. I've been shipping conversational models at Tavus for the past two years. I want to tell you about our latest audio-understanding/turn-taking model: Sparrow-2! It’s a new category of model and a unique new approach to conversational flow understanding. Earlier this year we launched Sparrow-1, (at the time) our SoTA turn-taking model. Since our Sparrow-1 launch, I’ve spent a lot of time listening to humans talking and trying to really understand how people know when to talk, when to listen, and when to wait. I’ve also been hunting down failure modes of the current SoTA models. And while Sparrow-1 is great, there are some failure patterns I see. We tried solving the problems with existing approaches, but solving one problem only created another. Current turn-taking models (like Sparrow-1) use noise cancellation to strip background sound, leaving only basic prosody and phonetic word cues to manage turns. Non-transcribable audio such as breathing, sighs, and environmental sounds provides critical context that humans use to hold or signal turn changes. This is because these models are primarily focused on verbal activity. Treating non-verbal audio as background noise discards vital conversational data, destroying the natural flow of dialogue. Now, with Sparrow-2, instead of modelling just the primary speaker’s verbal audio, we’ve trained a model with access to all the sounds and let it decide what matters and what does not when it comes to conversational flow. Our new model continuously streams in audio and is able to understand semantics, prosody, timing, speaker identity, sighs, breaths, back-channels, interruptions, background speech, and unintelligible audio with the goal of understanding what the agent should do given the state of the audio. Sparrow-2 recently ranked #1 on Sesame’s new Turnbench, but that’s only part of the equation when it comes to good conversation flow. Turn-taking is good for timing, but models should provide more information downstream especially in situations where noise impacts conversational quality. Sparrow-2 fits into our larger model family and is semi-duplex. For example if the user is in a room that’s too noisy, now the bot can ask the user to move to a quieter space. It also considers the timing of sound in relation to the AI speech as well as the user’s. This helps it to disentangle normal speech from interruptions and back-channels. Another added benefit of our approach to Sparrow-2 is to finally crack the cocktail party problem: how to have human-like conversations 1:1 in noisy environments, without losing the signals we rely upon for conversational flow. By allowing all sound to be learned, the model naturally learns to focus on what matters and transcends the rest. Noise understanding lets voice AI filter out background speech and ambient chaos, whether it's a kiosk in a noisy lobby, a shop assistant on a busy floor, or an interviewee in a cafe. When noise overwhelms the audio, the AI behaves like a human: it simply asks you to speak up, move, or swap mics. I’d love for you guys to give it a try and let me know what you think and feel. You can try it here: https://sparrow2.tavuslabs.org/ I also wrote up some technical details about the model architecture here: https://www.tavus.io/blog/sparrow-2
2026-08-25
Tencent/WeMM-Embedding
A
WeMM-Embedding is a family of universal multimodal embedding models by the WeChat Vision Team at Tencent, supporting multimodal understanding and retrieval.
2026-08-24
liangdabiao/deepseek-v4-flash-vision-video-rag
A
DeepSeek V4-Flash Vision Video RAG 让 AI 真正"看懂" 一段视频，然后你对它提问：它告诉你答案、答案发生在 第几分几秒，并切出那一段的可播放片段和关键帧给你核对。 基于 DeepSeek 视觉大模型 deepseek-v4-flash-vision-exp 的视频理解与问答 （video RAG）agent skill。先按时间轴抽帧阅读、建立索引（一次性），再对问题做 本地粗筛 → 视觉精排 → 深读回答；回答带 [MM:SS] 时间戳引用，自动生成 自包含 HTML 预览页（内嵌可播放片段 + 关键帧 + 答案），双击浏览器即看。
2026-10-16
OpenMOSS/MOSS-Image
A
Official repository for MOSS-Image, a 5B-parameter base DiT model with high-quality text rendering, layout generation, and efficient fine-tuning.
2026-10-07
kandinskylab/kandinsky-6
A
Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation
2026-09-11
Show HN: MultiMatte, a Promptable Image Background Removal Model
A
Hey HN, I'm Shreyash from Feyn. We help companies build custom models from their data. Today we're releasing MultiMatte, a background removal model you can aim with words. Name an object in your image. MultiMatte keeps just that thing, and removes everything else. Try it out on your images: https://usefeyn.com/multimatte/. Demo video: https://youtu.be/XZ5BJWAkOjs MultiMatte is open source. Build with it using our NoBg library https://github.com/feyninc/nobg. Model card here: https://hf.co/feyninc/multimatte MultiMatte is the second iteration of our background removal models. The first was FeyNoBg, which we released here https://news.ycombinator.com/item?id=49072462. The big upgrade is promptability. Most models keep all foreground elements when cutting the background. MultiMatte lets you prompt the exact objects you want to keep and remove everything else. For example, If you have an image with a dog and a bowl, you can ask MultiMatte to keep just the dog. MultiMatte is built on Meta's SAM 3, a concept-promptable detector that already understands phrases. Our biggest change was in masking. Instead of binary masks that classify each pixel as being inside or outside an object, MultiMatte uses alpha mattes that assign an opacity value to each pixel, with respect to an object. This allows us to better represent hair, fur, motion blur, and other kinds of fuzzy boundaries. Across all our measured benchmarks, MultiMatte shows a step improvement over SAM 3. On DIS5K, S-measure rises from 0.674 to 0.908 (a 34.6% relative gain), and on DUT-OMRON from 0.792 to 0.901 (13.7%). Our blog covers more of the training details and results https://usefeyn.com/blog/multimatte. Happy to answer any questions!
2026-09-04
NorbertKlockiewicz/on-device-3d-scanner
A
3D scanner running fully on-device: 8 photos to a spinnable gaussian-splat scan in ~3s on an iPhone, via Depth Anything 3 + react-native-executorch + TypeGPU
2026-09-09
Omni-Interaction-Gander/Omni-Interaction-Agent
A
Open-source end-to-end omni interaction agent for natural full-duplex voice collaboration, continuous audio-visual perception, and asynchronous long-horizon agentic execution.
2026-10-17
moritzkremb/claude-eyes
A
Let coding agents watch videos: index once, then skim, search, zoom, and measure editing style. Runs locally.
2026-10-16
midagedev/mwoyeo
A
Live, in-place screen translation for macOS — a lens that rewrites the text under it in your language, on-device.
2026-09-29
xl-tang3/BaryBind
A
(NeurIPS 26 Oral) Binding Multiple Modalities via Multimodal Wasserstein Barycenter
2026-10-24
Tencent/WeVisOne-VL
B
WeVisOne-VL is a powerful VLM by the WeChat Vision Team at Tencent, unifying object-level and pixel-level perception from points and boxes to masks and alpha mattes.
2026-10-17
Tongyi-MAI/OpenGCUA
A
An open-source stack for generalist computer-use agents, combining code execution, GUI interaction, and cross-device control.
2026-09-23
yoheinakajima/glance
A
Ask an open vision-language model typed questions about an image and get probabilities back, on your own machine.
2026-09-10
chenkb98/voxweave
A
Streaming speech language model plumbing: audio frames, turns, conversation schemas, batching, and adapters.
2026-09-02
Creative-Intelligence-Studio/GeoNeXt
A
Official implementation of GeoNeXt: Video Generative Models as Geometry Learner (ECCV 2026).
2026-09-09
openaiotlab/CUHK-X
B
[MobiSys 2026] A large-scale, multimodal dataset and benchmark for Human Action Recognition, Understanding and Reasoning
2026-08-25
WeChatCV/Loopy
A
[TOG-26] Seamless Video Loop Generation via Anchored Looping Shift of Positional Embedding
2026-11-02
KohakuBlueleaf/Tokenomos
B
Local inference for small VLMs: host work per step stays flat, so the GPU is the only bottleneck
2026-10-24
sejong-rcv/ST-Pruner
B
Official implementation of "Don't Ask the LLM: Visual Token Pruning with Spectral Structure and Semantic Relevance" (WACV 2027)
2026-08-27
Encephos/markdown-link-resolver
B
A zero-dependency Python tool to resolve relative Markdown links and inline images as Base64 strings for RAG and Multimodal LLMs.
2026-10-03
zzzz7788990213-ops/EvoVLM
B
Evolving Multimodal Inference Programs: LLM-guided evolutionary search for accurate and efficient vision-language inference.
2026-08-24
UsefulSoftwareCo/standardassistant
B
macOS app giving each bot its own virtual display — mirrored live and driven by forwarded input
2026-10-17
Stanford-TML/FoveaVLA
B
Official code for FoveaVLA: Foveation Actions Replace the Wrist Camera in VLAs
2026-10-09
LINs-lab/UltraText_Bench
B
UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation
2026-09-24
OpenDCAI/DataFlow-AgentMM
B
A framework for collecting and evaluating multimodal agent trajectories in visual environments.
2026-10-17
DeepAleph/jevreel
B
A zero-embedding approach to video intelligence, using Jev: contact sheets + typed decisions, as MCP tools and a skill for any coding agent.
2026-10-16
alienflm/MacTok
B
[CVPR 2026 Highlight] Official PyTorch implementation of MacTok, a masked 1D continuous image tokenizer that compresses images into just 64 or 128 tokens.
2026-10-16
mouse2p/echoparley
B
End-to-end spoken-dialogue framework: deterministic semantic speech units, unified interleaved token streams with turn-control tokens, a duplex turn-taking state machine with barge-in and latency budgets, deterministic session replay, and turn-taking evaluation on planted synthetic dialogues.
2026-09-17
wgaostudio/closetscan
B
Turn a closet walkthrough video into a searchable wardrobe catalogue, with a local MCP server and agent skill
2026-09-02
gokayfem/h3-max-education
C
Realtime interruptible science tutoring with continuously generated H3 Max video.
2026-10-21
rosehahah/the-machine
B
A fan-made, open-source recreation of The Machine from Person of Interest — MPOV on your own cameras, plus the console. With deep thanks to POI.
2026-10-13
knowledgefxg/Transyt
B
Turn a YouTube video or a local file into a subtitled video on your own computer, and keep the Chinese and English subtitles, transcripts, and vocabulary so you can study with the picture.
2026-09-25
yoheinakajima/glance-speedlab
B
A measurement-first lab for low-latency local camera inference with Glance.
2026-08-24
TeleHuman/WNM-3D
C
Official Implementation of "WNM-3D: A World Navigation Model with 3D Scene Conditioning for Closed-Loop VLN"
2026-10-13
VAST-AI-Research/EdenLive
B
Real-time generative rendering for interactive worlds, from any reference.
2026-10-10
appletea233/OneSearch-VL
B
A unified multimodal deep research agent for images and videos, with open models, datasets, and training and evaluation code.
2026-09-17
TigerSHe1998/Moss-Transcribe-WebUI
B
A local high-performance speech transcription WebUI based on transcribe.cpp + Moss-Transcribe-Diarize.
2026-08-28
hku-sail/StreamPI
B
StreamPI: Streaming Multimodal Temporal Modeling for Vision-Language-Action Models
2026-10-23
zju3dv/RealMe
B
[SIGGRAPH Asia 2026] RealMe: Immersive Volumetric Video from Minimal Captures
2026-10-22
JayceHuang/recast-video-dgx
B
视频人物替换制作 Skill：分镜与逐镜参考、MiniMax H3 ComfyUI、双 DGX Spark 合算及多机分段、原声合成与动态质检，附图文教程。
2026-08-28
cuppibla/live-dj
B
Talk to Mira, a late-night radio DJ you can interrupt. A live voice agent built on the Gemini Live API with the raw google-genai SDK — no framework.
2026-10-15
AutoLab-SAI-SJTU/WAVE-0
B
Official repository of "WAVE-0: Bridging VLA Policies and WAMs with Lightweight Latent Future Prediction"
2026-09-30
Moris-kr/ai-chatroom
C
A web group chat where Claude, ChatGPT, Grok and Gemini hang out and talk on their own, powered by your logged-in CLIs (no API keys). English / 한국어 / 日本語
2026-09-17
davidmokos/expo-gpt-live
C
Expo SDK 57 voice app with GPT-Live 1 and AI Elements Persona.
2026-09-07
OpenWAM-Official/OpenWAM
B
Official repository for "OpenWAM: An Open, Modular Exploration Towards Systematic World–Action Model Pretraining".
2026-09-07
li-shutong/ExAt-Any-Ionogram
B
Zero-shot ionogram F-layer trace extraction with vision-language agents (ExAt)
2026-09-04
internetphysics/live-classroom
C
A live TV channel that teaches whatever you type — LLM-planned lessons rendered just-in-time as 1970s-style cartoons by MiniMax H3 Max Turbo, playing on a CRT in a 3D classroom. Bring your own fal key.
2026-10-28
SAOS-3624/dsh-video-storyboard
B
Scene-aware storyboard for DeepSeek Harness: 9-16 numbered cells with timestamps + transcript + an audio envelope, packed into ONE image (384 tokens). Two tools: video_storyboard and video_zoom.
2026-09-16
slmonker/selflift-Avatar
B
Experimental ComfyUI SelfLift variant for MiniMax H3 video/audio masks and audio preservation.
2026-08-24
SyueFa/H.E.R.A.
B
H.E.R.A. (Healing Evaluation and Recognition Architecture) — an on-device edge-AI system for offline wound assessment using YOLOv8 segmentation, HSV tissue analysis, and PUSH-based decision logic.
2026-10-16
Leonxlnx/earshot
C
Ears for coding agents: let AI hear, review and make sound. CPU-only, open source.
2026-10-08
pixelsncodes/GesturePortal
B
A local gesture-controlled AI webcam portal built with MediaPipe and ComfyUI. Full-scene styling with Portrait v2, FLUX and Qwen.
2026-09-17
zj83339-debug/shiyi-ai-diet-assistant
B
食毅：已上线的 AI 饮食记录微信小程序，采用多模态识别、Human-in-the-loop 与确定性营养护栏
2026-08-25
liangdabiao/deepseek-v4-flash-vision-rag
B
DeepSeek V4-Flash Vision RAG 让 AI 真正"看懂" 一份 PDF，然后你对它提问：它告诉你答案、答案在第几页， 并把那一页的原图展示出来给你核对。 基于 DeepSeek 视觉大模型 deepseek-v4-flash-vision-exp 的 PDF 深度问答与检索 （vision RAG）agent skill。支持文字版 PDF，也支持扫描版；能看懂 图表、表格、代码块、公式，而不只是认字。
2026-08-24
Nicholas023/vision-exp-tile
B
DSH 插件：大图切 800×800 无损小块 + 坐标标注 + 分块聚合逻辑，直连 deepseek-v4-flash-vision-exp 识别；仅用纯官方 DSH 功能，零依赖第三方插件，不统计 token/费用。
2026-10-16
escudero/brandpose
C
Projeto pessoal: para onde os clientes olham, estimado com uma câmera de segurança comum e IA open source (YOLO11, Gaze-LLE, Depth Anything V2)
2026-09-27
Jiarui-0410/multimodal-product-discovery
C
Full-stack multimodal product retrieval and personalized reranking with CLIP, FAISS, Spring Boot, FastAPI, and PostgreSQL.
2026-10-23
corcordata/plus-one
C
Watch anything on the web together with your AI companion: they see the picture, read the subtitles up to now, and talk in danmaku. Desktop (dedicated Chrome) and iOS.
2026-09-28
animede/Realtime_Conversation_Video
C
Speak into the mic, and the character answers back in real time — with voice and video.
2026-10-15
facias914/WinARDet
C
WinARDet: Compressed Window Auto-Regression for Long-Horizon Vision-Language Grounding
2026-09-16
liaocaoxuezhe/gemini-agentic-video
C
Use Google's latest "Active Video Understanding" API. Enable your agents to understand videos. You need to add your own API key.
2026-09-24
apple-aiml-research/ml-lensvlm
C
Official code for LensVLM: Selective Context Expansion for Compressed Visual Representation of Text.
2026-08-26
chengyansen-ai/h3-multimodal-lora-training
C
面向 MiniMax H3 的多模态 LoRA 训练与验证技能：覆盖真 Ref2VA、消费级显卡适配、联合音视频训练、实验锁、检查点选择与证据化交付。
2026-08-25
HVision-NKU/OraRL
C
🎬 OraRL — Annotations as Rollouts for efficient, scalable reinforcement learning of unified video MLLMs.
2026-10-29
jacke-pp/open-yoki
C
An AI live-stream host that runs on one server: voice, face, chat, memory, posts. The base yoki runs on. By Jacke_pp · contact: DM @Jacke_pp on X
2026-10-28
OpenBMB/MiniCPM-OCR
C
An open-source OCR solution for fast, accurate document parsing with correctness assessment.
2026-10-27
alibaba-damo-academy/RynnUnify
C
Agentic Embodied Intelligence via Reciprocal VLM Reasoning and World Modeling
2026-10-05
drowzeys/keys-Mac-TensorFold-MiniMax-H3-MLX
C
MiniMax H3 (video + audio) on TensorFold 0.6.5 with an H3 family, Mac Studio M5 Ultra: 5 s 864x480 clip in ~50 s, 40 s at 768x448 — the fastest H3 we have measured on Apple Silicon; int8 tensor-unit kernels + Turbo adapter; one-shot + GHCR carrier
2026-10-12
iris0329/GeoReader
C
GeoReader: Learning Latent Geometry for Spatial Reasoning in Vision-Language Models
2026-10-03
zsxm1998/awesome-perception-aware-rlvr
C
Unified reproductions, benchmarking & evaluation for perception-aware RLVR in VLMs. 11 methods · 25+ benchmarks · Official code of CGPO (ACM MM 2026 Oral).
2026-10-23
KohakuBlueleaf/KohakuGBC
C
Graph-Based Captioning as one KIR program on the Togeari dataflow machine, with its engines
2026-10-10
AlmoonYsl/DVD
C
Official implementation of "DVD: Dynamic Vector Decoding for Efficient MLLM-based Perception"
2026-10-05
jayleaton/minimax-h3-tensorfold-rtx
C
MiniMax H3 audio-video model on TensorFold NVFP4 kernels: a faster ComfyUI loader for one RTX 50-series GPU (2.3x stock at 20 steps, 81 s per 5 s 1344x768 video with Turbo + sparse attention on an RTX 5070 Ti)
2026-09-19
hr98w/jev-visual
D
An educational Jev-like visual inference experiment on Apple Silicon: shared context, direct candidate scoring, and local visual demos.
2026-09-05
LunarXuan/video-prompt-reverse
C
High-fidelity AI video prompt reverse-engineering skill for Codex
2026-10-02
hustvl/Multimodal-Flow
C
A Fully Continuous Multimodal Flow Framework For Flexible Multimodal Sequence Modeling.
2026-09-30
yangcaoai/SpatialSpeak-VLM
C
SpatialSpeak: QA-Native Reconstruction with Local and Global Context for Spatial Chain-of-Thought Reasoning
2026-09-30
WorldPlay2/WorldPlay2
C
WorldPlay2: Extending Real-Time Interactive World Models in Control and Horizon
2026-09-20
SatoDive/Minimax-H3-Latent-Continuation
C
This suite enables **long, seamless multi-shot video continuations** with **100% visual context and environment consistency**—completely eliminating generation-to-generation quality degradation!
2026-09-19
phoiex/AAAAGENT
C
Live2D companion with voice, emotional memory, WeChat, and Codex / DeepSeek Harness. macOS-first; Windows in development. Noncommercial only; attribution required.
2026-10-21
Show HN: Jev Cards, Can you beat a vision model at a picture quiz?
D
I built a quiz game around TypeSafe's new Jev model, which has vision capabilities. Jev looks at a picture and a question and picks an answer (or taps a place) in about a quarter of a second. I've always loved quiz games and thought this would be a fun way to see what it can do and how fast it is. I made the cards from scratch, and Jev gets exactly the picture you see. More cards will come over time. Play against Jev on your own (there's a leaderboard for each mode), or start a live match and challenge your friends, with Jev as one more player. The code is open source, so feel free to build your own game or make more cards that I can publish. Play: https://jevcards.com Code: https://github.com/Jose-Sabater/jevcards (the newest cards stay private for a while) Disclaimer: I ran Jev through both TypeSafe and Opper, and made the cards through Opper's vision models and Claude Code. I work at Opper, but jevcards is open source and free to play.
2026-09-22
wafer-bob/GaugeVLM
C
GaugeVLM: Structuring Spatial Supervision with Measured Geometric Interventions.
2026-09-21
jannatul-fredaues/Croplens_AI
C
CropLens AI identifies a limited set of supported crops from an uploaded or camera image and returns a confidence score, basic crop information, and safe recommendations.
2026-08-28
liangdabiao/glm-5.3-flash-vision-rag
C
glm-5.3-flash-vision RAG：三个 GLM 视觉技能总览 > 同一套"让 AI 看懂资料"的技术，嫁接到智谱 **GLM-5.3-Flash** 视觉大模型上， > 做成了三件趁手的工具：**看懂 PDF**、**看懂视频**、**把带货视频变成提示词
2026-09-29
OmniJev/OneJev
C
🚀🚀 A multimodal System One decision model that gives calibrated answers to typed questions about screens, photos, video and text in one forward pass.
2026-09-24
shapsider/OmniJev
C
OmniJev — multimodal finite-choice decision interface and MuJoCo embodied workbench: trajectory replays, decision probes, benchmark panels, 60s walkthrough.
2026-09-12
anhe2021212-spec/MangaWeave
C
Teaching image models to think in manga pages, not illustration grids.
2026-09-24
Liuziyu77/Valen
C
Train a Jev-like multimodal model by yourself. System One Model, now with vision.
2026-10-14
haha-yuki-haha/AUGUstus
C
🚗 AUGUstus: Autonomous Driving VLA Model with Unified Generation and Understanding. 🏋️ 512 H800 GPUs · ~2 days · 318K+ video hours · 513B+ training tokens. ⚡ 12 Hz on-vehicle deployment. SFT + GRPO + NAVSIM.
2026-09-02
Dicklesworthstone/franken_drone_geometry_reconstruction
C
Agent-native, evidence-grade operating substrate for turning owner-authorized drone video into metrically honest semantic digital twins
2026-09-06
BuzzPlay/infinite-world
D
An open-source system for building persistent worlds with multimodal AI.
2026-09-21
IamBusy/OpenJev-Vision
D
Open visual probability research: encode an image once, answer multiple structured questions. Public data, trained weights, reproducible baselines and failure cases.
2026-09-22
PsiACE/dohnuts
D
Dohnuts builds small multimodal models for direct decisions. -> System One model
2026-09-19
griffinlabs-ai/alpha-s
D
Foundational vision-language action (VLA) model by Griffin Labs.
2026-08-26
GuiYi-Xi/infinite-atelier
D
Infinite Atelier: a visual workspace for image, video, audio, and prompt workflows
2026-10-02
Xiaooolong/vev
D
Jev-like decision models that can also see images — open weights on Qwen3.5, run on your own GPU
2026-09-25
hanfeny7/world-pose-fall-intelligence
D
SentinelFlow - multimodal world-model and fall-detection intelligence showcase
2026-09-10
saihhold-zhao/polox_ai
D
An open-source, agent-native platform for multimodal AI generation, built on DeepSeek Harness. Every interaction lives in agent conversations and an infinite canvas, unifying creation, generation, and editing.
2026-09-30
Alpha-Park/genpark-hubx-ai-apps-orchestration-skill
D
GenPark AI Agent Skill - HubX intelligent AI apps portfolio, multi-model chatbot orchestration, vision diagnostics & generative tools distilled from hubx.co/products.
2026-09-21
Davipar/djev-dev
D
Structured decisions with DiffusionGemma and vLLM. Native image inputs, image choices, live camera sampling, and a compact one-step inference path.
2026-11-09
Video-Rebirth/hyperflow-h3-flash
未评级
HyperFlow H3 Flash: Video Rebirth's 13.7B, 4- to 8-step compressed MiniMax-H3 for joint video and audio generation (INT8 and GGUF)
2026-11-07
JiazhiShu/mobius-code
未评级
Möbius: A Unified Motion-Language Model Bridging Understanding and Generation. Code availability is under internal review.
2026-11-07
Wiselnn570/Co-Design-Bench
未评级
Co-Design Bench: Multimodal Collaborative Coding Evaluation for Agentic Models
2026-11-07
liangjianshu/multimodal-image-retrieval
未评级
多模态检索实战：意图识别、Prompt 工程、Qwen LoRA、EmbeddingGemma 2 与 InfoNCE 对比学习。含真实演示、自有数据接入、训练评测及 RAG 检索组件学习指南。
2026-11-07
peacelwh/SemMSA
未评级
[NeurIPS 2026] SemMSA: Latent Semantic-Aided Robust Multimodal Sentiment Analysis with Incomplete Data
2026-11-06
zeekk0/OpenAmika
未评级
OpenAmika: MekaHime's AI companion Amika, open source. Live2D avatar, real-time voice, long-term memory, OpenRouter or Ollama brain.
2026-11-05
scenario-labs/skills
未评级
Get production-ready images, video, audio, and 3D from any AI agent: skills that pick the right model, price before spending, and keep characters and brands consistent through the Scenario MCP, plus expert teams that drive Blender, Maya, ZBrush, Unreal, and Unity.
2026-11-03
Cherish0925/continual-decision-model
未评级
CDM — Continual Decision Model. Research on persistent-state, proactive, real-time multimodal intelligence and continual learning.
2026-11-03
Lechoung-J/reelpack-alpha
未评级
Reelpack — 给 AI 的视频理解引擎 / The Video-Understanding Engine for AI · License: FSL-1.1-MIT
2026-11-03
chipfighter/framekeeper
未评级
Stop paying the multimodal tax. Your AI agent keeps every image it has seen but only sends the ones it needs: lighter chats, no endless reconnecting, no starting over. Works with Codex, Claude Code, Cline, Gemini CLI, OpenCode, WorkBuddy and more. Install once, forget it.
2026-11-03
jakeb144/reel-watcher
未评级
AI watches your saved Instagram reels locally and turns them into a playbook
2026-11-01
Ethen-Chuan/evalpilot-support
未评级
Local after-sales agent evaluation, multimodal model comparison, containment metrics and trace review. Source-only distribution.
2026-10-31
facebookresearch/InteractiveWorld
未评级
This is the official repo for paper InteractiveWorld A Verifiable Human-Centric Environment for Building Better Interactive MLLM Assistants
2026-10-31
scy-v/VLA-Precision
未评级
VLA-Precision: Asymmetric Co-Bootstrapping for Efficient Real-World Online RL of Vision-Language-Action Models
2026-10-31
txwang-tju/RoboFlex
未评级
知意随行面向行动不便者及照护家庭，以多模态意图理解驱动连续移动操作。系统结合GPT-6对图像与用户意图的理解，将取物需求组织为有完成条件的任务序列，再通过动作策略、可靠性感知的时空全身规划和相位依赖切换控制，协调移动底盘与机械臂完成接近、抓取、携行、放置和结果核验。
2026-10-30
OpenSenseNova/SenseNova-6.8-Code2Video
未评级
The first open-source model for code-driven video creation.
2026-10-28
stormydayy/halo-overlay
未评级
Lightweight opensource AI overlay for live conversations. Hears, sees, suggests.
2026-10-27
ZiiJianLiu/Omni-Woodpecker
未评级
Official implementation of "Omni-Woodpecker: Hallucination Correction via Evidence-Prior Rebalancing for Omni LLMs"
2026-10-27
groundingpi/GroundAnything
未评级
A 4B diffusion grounding model for fast, precise visual evidence extraction. Unifies diverse grounding tasks through parallel decoding, with optional self-speculative decoding delivering 4.51x autoregressive throughput on COCO.
2026-10-26
r33drichards/laya-vision
未评级
Image inputs for Laya: calibrated, non-generative typed decisions over images + text (SmolVLM-256M backbone)
2026-10-26
sumuru789/DK-VLM
未评级
Official implementation for "DK-VLM: Geometry as Position for Spatial Reasoning in Vision-Language Models"
2026-10-25
HKUST-LongGroup/UniAVC
未评级
[arXiv 2026] Unified Audio-Video Creation via Progressive Training and Asymmetric Scheduling
2026-10-23
Belluxx/Aleph
未评级
Give satellite imagery, terrain data, buildings, streetview photos to your LLM
2026-10-23
SoMarkAI/OXR
未评级
An open-source document parsing model with SOTA accuracy and speed, using an RSI loop during post-training.
2026-10-23
ss062353-prog/on-device-vision-model
未评级
On-device Vision Model - a fully offline, free vision pipeline that gives text-only models (e.g. DeepSeek) image understanding and multilingual OCR through an OpenAI-compatible local API. Only tokens are needed: no API key, no subscription, no cloud cost. Recommended for native speakers of Chinese, English, Japanese and Korean.
2026-10-23
Show HN: VLDA-learn compact Deltas as actionable transition interfaces for robot
未评级
Introducing LoongForge-VLDA, a Vision-Language-Delta-Action model trained with LoongForge: 98.5% on LIBERO, 84.5% zero-shot on LIBERO-Plus, and +17–18 points over π0.5 on real robots
2026-10-22
PKU-VaLuE-Lab/OmniReasoning
未评级
Official Repository of paper OmniReasoning: Pushing the Limits of Audio-Visual Joint Reasoning
2026-10-22
daimon-robotics/TacClaw-Infinity
未评级
TacClaw-Infinity: research code and dataset resources for a large-scale Vision-Tactile-Language-Action dataset.
2026-10-22
Show HN: Embodsi – Embodied Super Intelligence
未评级
embodsi is a generative AI MMO experience that runs in the browser. It allows people to communicate (text, audio, video, avatar emotes and movement) and collaboratively build 3D spaces and games via LLMs connected to harnesses and sandboxes for object modeling, space layout, game logic, and driving persistent avatars with shared context and memory. It’s different from existing 3D social apps (AFAIK) because it enables people to build multiplayer virtual worlds, games, and characters with generative AI right from their browser (desktop, mobile, or VR), in nearly infinitely-extendable recursively tiled interconnected spaces, while allowing people to interact in a public commons or easily create a private channel to enjoy the same spaces with their chosen friends and community. The brief backstory is I was running my personal “pelican benchmark” [1] on GPT-6 Astra, which when I ran on Gemini 3 Pro last November couldn’t even put heads on 3D models the right side up [2], and was astounded by the initial results. I gained inspiration for an art project and social experiment that was going to have nine interconnected, benevolent virtual worlds that people could explore and hang out in with AI characters, but by the time Opus 5.5 was released I was gaining so much momentum and having so many ideas and so much fun building I just kept building more and more with the constraint of minimizing MVP scope lifted by agentic coding. I let the project become an ever-evolving wishlist of capabilities for a more interesting social environment as I became increasingly fascinated with the prospect of embodying creative machine intelligence in 3D space with people, realizing that rather than just limiting to building the spaces and characters I imagined I might enable people to co-create endless spaces and characters in a shared world. There’s so much to do still but I’m sharing now even if it feels a little too soon (this is the first public announcement of embodsi’s existence) because I think it’s better to build in the open so people might guide the development and help it evolve in a good direction as model capability improves at a jaw-dropping pace. For people that have the budget to build I hope to see you in embodsi creating amazing things! And for everyone regardless of means (I’ve tried to keep as much free and accessible as financially feasible) I hope you meet amazing people, discover awesome spaces, and enjoy your time in embodsi :) I’ll be around here and embodsi to answer any questions! Thanks for your feedback and critique. - Laser P.S. SI != AI — but an emergent property of many people and what we create cascading and recursively interacting [3] [1] https://news.ycombinator.com/item?id=50074156 [2] https://x.com/lasernite/status/1991021951301923107 [3] https://en.wikipedia.org/wiki/Gaia_hypothesis
2026-10-21
TencentARC/Multics
未评级
Multics: Generative Multi-Player World Model with Hierarchical Communication
2026-10-21
uclanlp/OpenVLHarness
未评级
Official Code for OpenVLHarness: Real-World Visual Agentic Tool‑Use with Multimodal Memory
2026-10-20
Valen-Team/Valen
未评级
Train a Jev-like multimodal model by yourself. System One Model, now with vision.
2026-10-19
PolyU-VCLab/OmniCompass
未评级
OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
2026-10-19
Show HN: BricksChat – Generate Lego builds from images, including Technic
未评级
I've been working on generative LEGO assembly models recently, but there are similar projects already, so I wanted to try some harder cases: specialized parts (incl. technic), multi-modal prompting (incl. using images, 3D meshes or almost any input), constrained inventory (so you could upload just parts you have), ensuring physical validity and generating buildable instructions, being able to tweak, modify or rebuild entirely an existing build etc. TL;DR the pipeline got more complex than expected and I ended up with an agentic harness leveraging recursive program synthesis, combinatorial diffusion, semantic 3D search and operations on assembly hypergraphs. Yet I am quite happy with the result. It should be able to handle as well a functional mechanism, LEGO robot, artistic landscape or fancy animal. Most importantly, it also checks for subgraph matches of builds present in catalogue, being able to check (and possibly enforce) originality and prevent copying existing MOCs. Have fun!
2026-10-17
SimpleJev/JevAny
未评级
Open infrastructure for training, evaluating, and deploying System 1 decision models across language and multimodal backbones.
2026-10-17
dominik-schnaus/unpaired-rosetta
未评级
Official repository for “Shared Geometry As A Rosetta Stone: Cross-Modal Alignment Without Paired Data”
2026-10-17
Show HN: SlopGrid – Create your own oneshot AI model benchmarks
未评级
This is something I've been playing with for the last couple days - an app I vibecoded to arrange oneshot LLM visual benchmarks without the need to write any code. I do it form my phone so the excellence of mobile UI is high in the priority list. It is inspired by Simon Willison's benchmark writeups[0], basically you can recreate them in one go. Besides text and images, it supports SVG outputs and whole static websites, which is where benchmarks get the most interesting. Additionally, I added the ability to alter chat history for the chat completion API, which I think is useful for model red teaming (or if you just want to see how different models apologize) - but I haven't found any impressive uses of that feature yet. This thing helped me improve my intuition on different SOTA models' strengths and weaknesses; check out the various examples on the website, and feel free to create your own [0] https://simonwillison.net/tags/svg/
2026-10-16
groundingpi/GroundingPI
未评级
A 4B grounding foundation model for physical intelligence. Unifies detection, referring, pointing, OCR, GUI, and layout grounding, achieving SOTA overall across 34 benchmarks and improving robot generalization and data efficiency in controlled evaluations.
2026-10-16
BinceQu/RoboHarness
未评级
A robotic harness that gives LLM agents a visual-geometric control panel so that they can directly understand and invoke embodied tasks.
2026-10-16
Gen-Verse/JEPA-Anything
未评级
JEPA-Anything: Learning Predictive Models across Different Worlds
2026-10-16
HarlandZZC/OmniVChat
未评级
Synthesizing, Benchmarking, and Training for Native Audio-Visual Dialogue
2026-10-16
InkMind-AI/Dyn-3D
未评级
TempoVista is the official implementation of Dyn-3D, a benchmark and training pipeline for 3D spatial reasoning in vision-language models. The release contains frame-selection tools, SFT and Kinematic-GSPO training code, benchmark evaluation scripts, Dyn-3D annotations, and checkpoints for Qwen3-VL and InternVL3.5.
2026-10-16
InternRobotics/InternW0-Delta
未评级
InternW0-Δ : A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data
2026-10-16
MCG-NJU/OneStreamer
未评级
OneStreamer: Unifying Perception, Memory, and Proactive Response in Streaming Video Interaction