Είναι σχεδόν βέβαιο. Το κάθε πρόσωπο πλέον υπάρχει παντού (κάμερες ασφαλείας, τυχαίο ζευγάρι που φωτιγραφίζεται μπροστά σου κτλ). Εταιρείες αγοράζουν όσο-όσο τέτοια δεδομένα. Δεν χάνεσαι. Πρακτικά 20 κεντρικές κάμερες με insightface και σε βρήκε. Εκτός αν την έχουν Κωσταλέξι.

Fooocus is an AI image generator for creators and designers, offering advanced inpainting, multi‑prompt support, style controls and InsightFace‑based face swapping to turn prompts into high‑quality visuals instantly. fooocus.one/ #TexttoImage #PhotoImageEditor

Codex的回答 不是 MCP，也不是外部 API。 这次换脸是用你本机 ComfyUI 里的本地 ReActor 节点模型跑的，具体是： 本地模型：models/insightface/inswapper_128.onnx 本地脚本：我临时写了一个批处理脚本逐帧换脸 合成：用本机 ffmpeg 把换脸后的视频和原音频合回去 我一开始检查过 LibTV/API 路线，但你这边没有配置 LIBTV_ACCESS_KEY，所以没有调用外部生成 API。视频处理全程在本地完成。

codex solito ha descargado las fotos de flicker (ni idea de como), ha buscado en la documentación de varios repos de object detection y ha decidido que InsightFace era la mejor opción, ha instalado las librerías, descargado los modelos y ha escrito el pipeline para detectar las caras en todas las fotos y calcular el cosine similarity con los crops de las fotos de Leire y me ha dado los resultados literalmente no he hecho nada yo

nah, parece que no sale en ninguna foto de aquí estas son las caras con mayor cosine similarity al vector embedding de Leire y nah aunque ojú, algunas es complicado decir lmao btw: codex ha usado InsightFace (buffalo s, en CPU) no es SOTA pero dudo que se le haya pasado

Bu iş nereye gidiyor yahu… 😂 Başta “İki tane ESP32-CAM bağlayayım, YOLO ile insan görsün, gözü gönlü açılır, biraz insanlık görsün.” dedim. Şimdi elimde: GTI9 Pro L GTI13L ile 2 kamera, YOLO26 InsightFace (Face Rec Re-ID), Qwen 3.6 Ollama havuzu akıllı dispatcher, NATS JetStream, Qdrant, SeaweedFS, Timescale, React dashboard… Kısacası: ESP32’den başladığım proje kurumsal seviyede, 300 kameraya ölçeklenebilir, tamamen yerel çalışan bir AI izleme platformuna dönüştü. Konu ESP32’ydi… iş çığrından çıktı ama çok güzel çıktı. WatchLLM artık resmen kendi kendine evrilen bir canavar oldu 😄 #WatchLLM #LocalAI #ESP32 #YOLO #Qwen #BuildInPublic #NeYaptımBen

This guy makes $6,800 a month from a single skincare brand because he built an AI influencer that shoots their ads instead of booking real models. And that is one brand. He runs several at once, each on its own monthly retainer. The influencer is a woman who does not exist. He generated her once, and now she models whatever product a brand sends him, in any outfit and any setting, with no studio, no photographer, and no model day rate. When a brand needs her talking on camera, he films himself and an AI layer paints her over his face live, frame by frame, exactly like the clip going around where a guy in a hoodie turns into her in real time. Here is why D2C brands actually pay for this. A beauty or clothing label needs fresh content every single week. A real shoot means booking a model, a photographer, and a studio, then waiting days and paying $300 to $800 for one usable clip. His AI model puts out as many shots as you want, with the exact same face every time, in a day, for a flat monthly fee. She never ages, never reschedules, never shows up with a breakout, and never drags the brand into a scandal. Virtual faces also pull around three times the engagement of human creators, so the brand often gets cheaper content that performs better. He generates one consistent character with a diffusion model, the same Stable Diffusion or Flux you already know, until her face is identical in every shot. A real-time face swap model, the open-source one built on InsightFace that powers tools like Deep-Live-Cam, maps that single image onto his webcam with no training and runs on his own GPU. A cloned voice handles the audio, and an LLM writes her captions and replies. One source image, one swap model, a few scripts. The money scales the boring way. One brand is $6,800 a month, a handful of brands is a real income, and the next step is a whole roster of characters. The top AI personas already pull $20,000 to $200,000 a month once they run several streams at once. The wildest part is the asymmetry. A real model shoots for one brand at a time and goes home. His AI model works for 10 brands at once, overnight, in 10 different styles, and never charges for overtime. One person now carries what used to take a studio, an agency, and a whole roster of faces. In a year this stops being a clever trick and becomes the norm. One question will be left: will you be the one building these, or the one paying for them. Which side will you end up on?

用一张照片加音频实时生成口型同步的 3D 头像视频，单卡就能跑 25 帧。。 github.com/avaturn-live/avtr… AVTR-1 是 avaturn-live 开的实时对话头像模型。给一张肖像图加两路音频，它就能逐帧生成口型同步和倾听表情，输出 25fps 视频。 支持单人和双人对话。A100 上 5 帧延迟 91ms，RTX 4060 Ti 上 166ms，实时够用。权重从 HuggingFace 下，推理用的 InsightFace 的人脸检测和关键点模型。

That a made me laugh!! 😂🤣. Dyslexics unite! The Reachy, with it's current "brain" isn't going to run much more than a Qwen3.5 2B Q4 model or similar, I did try the Granite4.1-3B for tool calling but it kill everything else (in parallel). This is partly why I wanted to upgrade it, my goal was a 100% local robot. I can do that but it's limited, for now it uses TailScale to partner with one of my Mac Minis and that uses ChatterBox-Turbo and Granite4.1, SAM3.1, insightface & Qwen3.6 35B. With that it's at least private.

I’m currently leveraging the InsightFace library to build a streamlined facial surveillance system!🥂 The pipeline is quite simple as at now: >load a target image -> detected once >image got embedded -> stored in memory >video -> processed frame-by-frame -> each frame's faces get detected -> each detected face gets embedded -> compare frame embedding with stored target embedding -> draw a bounding box on detected target's face -> discard frame embedding after comparison Eager to share more details as it progress! 🥂🫡

ComfyUI向け高速・クリーンな顔交換カスタムノードが公開。FLUXとInsightFaceを活用し、顔の切り抜き、マスク生成、参照latent条件付けに対応。低品質画像の顔一貫性向上も実現し、後処理や比率ヘルパーも付属。高品質入力でリアルな結果に優れ、プロンプトによる柔軟な調整が可能。すぐ利用できる設計。 #ComfyUI #顔交換AI URLはリプ⬇️

ComfyUIの高速顔交換ワークフロー。FLUXとInsightFace統合のカスタムノード。 顔領域抽出、マスク生成、低品質画像の顔一貫性向上、色調整など後処理機能搭載。 高品質画像でリアルな結果。プロンプトで柔軟な調整も。CPUでも高速動作。 #StableDiffusion #ComfyUI URLはリプ⬇️

Frigate InsightFace a vector database Aqara U100! Unlocks the door for family without needing the key

🧬 **Yapay Zekanın Yeni Gerçeği: Tek Bir Fotoğrafla Gerçek Zamanlı Yüz Değiştirme** Dikkat çeken bir açık kaynak proje: **Deep-Live-Cam**. Sadece bir insan yüzü fotoğrafı yükleyerek kameranızda, canlı videoda ya da herhangi bir videoda anında yüz takası yapabiliyorsunuz. - Hiçbir post-prodüksiyon yok - Export beklemiyorsunuz - Gerçek zamanlı çalışıyor (webcam, OBS, Zoom, canlı yayın… her yerde) Proje GitHub’da **89 binden fazla yıldız** almış durumda (github.com/hacksider/Deep-Li…). InsightFace tabanlı modeller, GFPGAN yüz iyileştirme ve ONNX Runtime ile GPU/CPU desteği sunuyor. Multi-face desteği bile var. **Ancak işin öteki yüzü kritik:** Bu teknoloji dolandırıcılık, kimlik sahtekarlığı ve dezenformasyon için yüksek risk taşıyor. Geliştiriciler etik filtreler koymuş (NSFW içerik bloklanıyor) ve “kullanırken consent alın, çıktıyı deepfake olarak işaretleyin” diyorlar. Yine de 2026’da bu tür araçların kötüye kullanımı hızla artıyor. Kreatif tarafta ise VTuber’lar, meme üreticileri ve içerik yaratıcıları için devrim niteliğinde. Video’yu izleyin 👇 ve söyleyin: Bu teknolojiyi **etik sınırlar içinde** kullanmak mümkün mü, yoksa regülasyon şart mı? #DeepLiveCam #YapayZeka #Deepfake #Teknoloji #AI

2月に報告されたフィッシングキット「Starkiller」の作者とみられる脅威アクターが、AIディープフェイクでKYC（本人確認）を突破するツールをダークネット上で販売しているとの報道。「Jinkusu」と呼ばれるこのアクターは暗号資産取引所や銀行の本人確認プロセスを標的としたツールを提供しているとされ、オープンソース顔認識ライブラリ「InsightFace」によるリアルタイムの顔入れ替えと音声変調で生体認証を回避するとのこと。 【要点の整理】 ・同報道によると、InsightFaceによるリアルタイムの顔入れ替えと音声変調を組み合わせ、KYCの生体認証を回避可能との説明。 ・技術的知識がなくてもロマンス詐欺に転用可能とされ、同報道によると暗号資産関連の同詐欺は2024年だけで数十億ドル規模の被害が報告。 ・Jinkusuは2月にセキュリティ企業Abnormalが報告したリバースプロキシ（通信中継）型フィッシングキット「Starkiller」の作者と同一とみられる人物との指摘。同Abnormalの報告によると、Starkillerでは標的ブランドの本物のWebサイトがヘッドレスブラウザ（画面なしで動作するブラウザ）経由でリアルタイム中継され、MFA（多要素認証）も突破される構成とのこと。 ・ディープフェイクによるKYC突破は、2024年に報告されたツール「ProKYC」等で先行事例がある既知の手法。同一アクターがフィッシングとKYC突破の双方をサービスとして提供している点が焦点と報じられているもの。 cointelegraph.com/news/ai-cy…

Darknet'te "Jinkusu" takma adlı bir aktör, Binance, Coinbase, Kraken ve bankaların KYC sistemlerini gerçek zamanlı yüz değiştirme ve ses klonlama ile atlatan bir siber suç aracı satıyor. Araç açık kaynaklı InsightFace modelini kullanıyor. Canlı video görüşmesi sırasında yüzünüzü anında değiştiriyor, jestlerinizi ve dudak hareketlerinizi hedef kişiyle eşleştiriyor. Üstüne ses modülasyonu ekliyor arayan kişi sizin sesinizle konuşuyor, sizin yüzünüzle görünüyor. Tüm bunlar canlı, gerçek zamanlı çalışıyor. Teknik bilgiye gerek yok. Binance Güvenlik Direktörü Jimmy Su, Mayıs 2023'te uyarmıştı: "Gelişen AI algoritmaları kurbanın tek bir fotoğrafıyla KYC sistemlerini aşacak." Tahmini şimdi gerçek oldu. Çalınan bir selfie, birinin banka hesabınızı açması için yeterli. Jinkusu'nun aynı kişi, Şubat 2026'da Starkiller adlı phishing aracını da piyasaya sürdüğü şüpheleniliyor. Starkiller sahte giriş sayfası kullanmıyor; hedef markanın gerçek giriş sayfasını Docker içinde headless Chrome ile yükleyip tüm kullanıcı verilerini gerçek zamanlı saldırgana iletiyordu. Sadece kripto da değil. Araç "pig butchering" adı verilen romantizm dolandırıcılığında da kullanılıyor. 2024'te yalnızca kripto yatırımcıları bu tür 200.000 vakada 5,5 milyar dolar kaybetti. Cyvers CEO'su Deddy Lavid'ın uyarısı net: "AI sentetik kimlik dolandırıcılığının giriş engelini düşürdükçe, platformların ön kapısı her zaman savunmasız kalacak." Çözüm katmanlı güvenlik davranış analizi, cihaz parmak izi, gerçek zamanlı AI izleme. Kaynak: Dark Web Informer, Cointelegraph (6 Nisan 2026), TorNews

🚨 CRYPTO SECURITY ALERT: THE END OF FACIAL VERIFICATION (KYC) 🚨 🌐 The launch of JINKUSU CAM—a cybercriminal tool—has been detected. It is a powerful AI suite designed specifically to BREACH the security protocols of the world's largest exchanges (Binance, Coinbase, Kraken, OKX). 👤 Developer/Threat Actor: jinkusu. 🛠️ Tool Type: Real-time media manipulation software (Live Deepfake). 🎯 Objective: To bypass KYC (Know Your Customer) protocols on financial platforms, cryptocurrency exchanges, and mobile banking applications. 📦 TECHNICAL FEATURES (ATTACK VECTORS): The software utilizes cutting-edge AI technologies to deceive verification systems: 🎭 Real-time Face Swap: GPU-accelerated face replacement (CUDA/DirectML) using InsightFace for fluid gesture transfer. 🗣️ Voice Changer: Real-time voice modulation with pitch adjustments and preset profiles (Anonymous, Radio, Robot) to evade voice biometrics. 🎥 Virtual Camera: Output compatible with OBS Virtual Camera, allowing the manipulated video feed to be injected into Zoom, Teams, Chrome, and verification apps. 📱 Emulator Support: Designed to function within Android emulators, enabling attacks against mobile applications that require "live" selfies. ✨ AI Enhancement: Utilizes GFPGAN and 478-point facial meshes (MediaPipe) to ensure the fabricated face mimics human expressions with extreme precision. ⚠️ ASSOCIATED RISKS (CRITICAL): 🏦 Mass Banking Fraud: Enables criminals to use stolen photos (such as those from previous data leaks in France or Mexico) to create a "living" persona that speaks and moves convincingly before a bank's camera. 🎭 Synthetic Identity Theft: Facilitates the creation of highly convincing fake identities for money laundering activities and romance scams (pig butchering). 🔓 KYC Compromise: Bypasses the "Liveness Detection" checks that many applications consider secure. #Cybersecurity #Deepfake #KYC #Bypass #JinkusuCam #AI #IdentityTheft #Fintech #InfoSec #CyberAlert

要5090卡的，普通人就别想了。视频中的AI变脸技术主要是基于InsightFace人脸检测 扩散模型/GAN的面部交换（Face Swap），能做到近100%相似度。很可能用了FaceFusion这类开源工具（免费、视频处理快）。 自己实现超简单：去GitHub搜“facefusion”下载，录制自己视频，输入评论者照片作为源脸，一键生成。需要NVIDIA显卡，10分钟就能上手！