Qwen-Image-2.0

Qwen-Image-2.0 is Alibaba Cloud's second-generation image foundation model, released on February 10, 2026. It combines image generation and editing into a single model, replacing the separate pipelines required by earlier approaches. The architecture pairs an 8B Qwen3-VL encoder with a 7B diffusion decoder, producing a leaner design than its predecessor (which used 20B parameters) while achieving higher benchmark scores. It supports native 2048x2048 resolution output.

General
Release date	10 Feb 2026
Developer	Qwen / Alibaba Cloud
Type	Image generation and editing model
Architecture	8B Qwen3-VL encoder + 7B diffusion decoder
GitHub	QwenLM/Qwen-Image
Documentation	qwenlm.github.io

Core Features

Unified generation and editing: a single model handles both text-to-image generation and instruction-based image editing without switching pipelines.
Native 2K resolution: generates images at up to 2048x2048 pixels natively.
Professional typography: supports up to 1,000-token prompt instructions for text-heavy visuals, generating accurate text in posters, infographics, slides, and comics.
Photorealism: produces fine-grained detail at 2K including skin texture, fabric weave, and natural foliage.
Efficient architecture: 7B active parameters vs. 20B in Qwen-Image 1.0, with faster inference and higher quality.

Benchmarks

Benchmark	Qwen-Image-2.0	FLUX.1 (12B)
DPG-Bench	88.32	83.84
AI Arena (text-to-image)	#1
AI Arena (image editing)	#1

Tools and Resources

GitHub (QwenLM/Qwen-Image): model code and usage examples.
Qwen API Platform: API access for generation and editing endpoints.
Qwen Studio: web interface to try image generation and editing without setup.

Ecosystem and Integrations

Accessible via Alibaba Cloud DashScope API for programmatic generation and editing.
Available through Qwen Studio for no-code testing.
The encoder backbone (Qwen3-VL) is the same model used for vision-language understanding tasks.

To get started, visit Qwen Studio for a live demo, or use the Qwen API Platform for API access with the DashScope SDK.

RAG-MATCH3-AMD

RAG-Match3-AMD is an intelligent document question-answering system designed to help users find reliable information inside PDF, DOCX, spreadsheet, and text documents. The project uses Chroma Cloud as its vector database and retrieval engine. Documents are parsed, divided into manageable chunks, and transformed into dense embeddings with Chroma Cloud Qwen and sparse embeddings with Chroma Cloud SPLADE. Hybrid search and Reciprocal Rank Fusion combine semantic and keyword results to retrieve the most relevant evidence. OpenAI GPT-4.1 Mini then analyzes the retrieved context and generates a concise answer supported by citations to the original document. The system can also process images and scanned content through a vision-capable model. Document metadata, chunk indexes, and source identifiers are preserved to improve traceability and prevent duplicate results. The project provides a practical RAG workflow that separates document retrieval from answer generation. Chroma Cloud handles embedding and search, while OpenAI handles language understanding and response generation. This architecture makes the application suitable for learning materials, technical documentation, reports, and other knowledge-intensive document collections

Optiseek

Mini Challenge 2: OCR Image: b3panda3/optiseek:latest Optiseek extracts text from PNG/JPEG/TIFF license plates and traffic signs using Qwen2.5-VL-3B-Instruct on AMD ROCm, returning strict JSON with text and confidence. Categories: US/Chinese plates, stop signs, speed limits, advisory speeds, and work zone signs. Preserves relevant text while filtering US state names and slogans. Architecture: PIL adaptive preprocessing → Qwen2.5-VL (bf16) → rule-based postprocessing → JSON. Results: 10/10 samples passed (200/200 points); 17.2s/image average; ~8 GiB peak VRAM; ~25 GiB Docker image. Engineering: Patched Transformers caching_allocator_warmup for MI300X VF memory limits; used a 3B model for VF compatibility. Mini Challenge 3: RAG Image: b3panda3/optiseek-rag:latest Optiseek RAG indexes PDF, DOCX, XLSX, CSV, TXT, LOG, PY, PNG, and JPG files. It answers questions with exact citations and abstains when evidence is missing. VLM extraction handles text in diagrams, pinout charts, and asset labels. Robustly handles unreadable files, encrypted PDFs, empty directories, and unknown formats. Architecture: Resident Qwen2.5-7B daemon + Unix-socket client → multi-format parsers → BGE-small embeddings + hybrid semantic/keyword retrieval → LLM generation → post-hoc citation necessity verification. Results: 8/10 questions passed (160/200 points). Supports single-hop, multimodal, and refusal tasks. Multi-hop retrieval and one error-code extraction need improvement. Engineering: Daemon avoids model reloads; VLM unloads after indexing to free VRAM; citation verification tests necessity by removing sources and rerunning queries. Patched caching_allocator_warmup and pinned ROCm-compatible PyTorch to prevent CUDA replacement. ~50 GiB Docker image; ~14 GiB peak VRAM.

VoiceDesk AI

VoiceDesk AI is a real time voice support agent that feels like calling a helpdesk, except the agent can actually see what you see. A user opens the web app, starts a call, and talks to an AI assistant with a photorealistic avatar that speaks back with lip sync and sub second response times.The agent watches the user's shared screen so it understands the actual problem instead of relying on a vague description. When a user says their account is blocked, the agent reads the error on screen, confirms the issue, and calls a tool to unblock the account directly. Once the issue is resolved, it sends a ticket summary by email so there is a record of the session.Under the hood the agent runs as a LiveKit worker built from independently chosen best in class providers: AssemblyAI for streaming speech to text with end of turn detection, Groq running a vision capable Qwen model for fast reasoning over both speech and screen frames, Deepgram Aura for low latency text to speech, Silero for voice activity detection, and Beyond Presence for the avatar. The frontend is a Next.js app that publishes the user's microphone and screen and renders the avatar in the browser.VoiceDesk AI shows what a modular voice pipeline can do beyond simple question answering: a support agent that sees, listens, reasons, and takes real action on a user's behalf.

Fresno LUX: A Living City

Fresno LUX is a prototype for a living intelligence layer across an entire city. Fresno already contains enormous amounts of intelligence: residents know where problems exist, nonprofits know where help is needed, businesses know where opportunities exist, educators understand gaps in learning, researchers generate knowledge, and government maintains critical infrastructure and data. The problem is that much of this intelligence is fragmented across disconnected systems. Fresno LUX connects those pieces. The platform creates a shared city graph where people, organizations, resources, needs, projects, observations, experiments, and outcomes become connected nodes. Instead of simply collecting data, LUX continuously asks: What is happening? What do we know? What is missing? What can we test? What changed? A community problem can become an observable signal. That signal can become an experiment. The experiment produces evidence. Evidence changes the system's understanding. That updated understanding informs the next action. Observe → Connect → Understand → Act → Measure → Learn → Converge. The goal is to move Fresno from disconnected information toward collective intelligence—a system where knowledge can travel across boundaries and successful solutions can be discovered, tested, measured, and replicated. LUX is designed to make the invisible connections visible: the relationship between a neighborhood and its resources, between a student and educational opportunity, between a business and workforce capacity, between environmental conditions and community health, and between an idea and measurable real-world impact. This hackathon prototype demonstrates the beginning of a Living Fresno Graph: a city that can observe itself, learn from evidence, and coordinate toward better outcomes.

SolInfinite Alpha

SolInfinite Alpha is a production-grade autonomous multi-asset trading ecosystem engineered for the Alpaca AI Trading Agents Hackathon. The platform bridges institutional-grade quantitative risk modeling with next-generation artificial intelligence to execute automated strategies across both US equities, options chains, and digital assets. At its architectural core, a high-performance Flask backend leverages the official alpaca-py SDK for live multi-market order routing alongside the Google GenAI SDK to continuously process real-time market regimes, implied volatility ranks, and core quantitative metrics. When market conditions align with optimal parameters across supported stocks, options, and cryptocurrencies, the autonomous agent evaluates confidence thresholds exceeding 80% to programmatically deploy defined-risk credit spreads, equity positions, and crypto orders directly into the live Alpaca execution pipeline. Designed with an immersive dark-mode glassmorphism interface styled via Tailwind CSS, the platform delivers interactive fluid cursor physics, real-time telemetry tracking cross-asset portfolio equity, cash, and buying power, alongside an integrated AI chatbot gateway. This assistant provides granular chart breakdowns, market situation assessments, and structured educational modules, ensuring a seamless balance between automated algorithmic execution and comprehensive cross-market analytics.

LEFA AI - Governed Autonomous Alpha Companion

Most finance products introduce themselves with charts, balances, and buttons. You speak to LEFA. LEFA is the user-facing base intelligence. The complicated architecture underneath exists to help LEFA make a better decision — not to make the user learn the architecture. You bring the human question. LEFA brings governed financial intelligence. See the engine room Inspect assets. 👁️ Observe → 📖 Ledger → ✨ Reveal Animated LEFA Observe Ledger Time Reveal timeline LEFA does not treat a convincing first answer as reality. Observe what is happening. Ledger what was known. Let time continue. Reveal what survived. The frontend tells the story. The backend preserves the truth. Time decides what survives. 🧠 What moves under LEFA? Animated LEFA backend control room The public experience stays light. The backend is intentionally heavy. HUMAN ↓ LEFA — base identity / final decision-maker ↓ CRUD — capture the event ↓ ARK — governed structured bloat ↓ BMP—stress test, filter and compress ↓ MAO — route bounded responsibility ↓ SWFUS — five internal ecosystems ↓ EVIDENCE + RECEIPTS + UNCERTAINTY ↓ LEFA ↓ DECISION / EXPLANATION The internal system does not replace LEFA's judgment. It filters unsupported certainty, preserves useful ambiguity, and returns better governed evidence to the base model. OPEN // Why the backend expands before it compresses 🌱 Five agents. Five ecosystems. One LEFA. Animated LEFA SWFUS five-agent ecosystem The five internal agents do not talk to the user. They receive governed information from LEFA's backend and operate inside bounded ecosystems. S Sovereign Ingestion What may enter as a governed signal? W Witness Isolation What must be preserved independently as testimony? F Fluid Vectoring What interpretations or directions remain plausible? U Unified Synchronization What accepted state must align before action? S Severance Execution What must be cut, held, or rejected? Rich at the center. Minimal at the edges.