分类筛选
找到 748 个 Demo
Nano Banana Pro which is built on the new Gemini 3...
Nano Banana Pro which is built on the new Gemini 3 Pro Architecture is real and it is a massive leap forward. It moves beyond just "making pretty pictures" to visual reasoning, meaning it understands complex instructions, accurate text rendering and real-world data integration better than any previous Google model. To make Nano Banana Pro a success, you must shift your prompting strategy from descriptive (e.g., "draw a cat") to structural. • Why: Unlike older models that just "hallucinated" pixels based on keywords, Nano Banana Pro has a "thinking" stage (similar to reasoning models). It plans the layout before it renders. • The Fix: Your prompts should explicitly define Layout, Typography, and Camera Telemetry. • My personal take: Don't overload one prompt. Instead, use a "Chain of Thought" approach. First, ask it to plan the image (e.g., "Generate a text description of a movie poster layout"), and then feed that plan back into the image generator. This leverages its new reasoning capabilities for maximum precision. Treat the AI as an Art Director, not a Painter Here are 10 prompts designed to stress-test the specific new capabilities of Nano Banana Pro: Text Rendering, Complex Reasoning, World Knowledge and Technical Photography. Thread🧵👇🏾
built FrameForge at Gemini 3 Hackathon 🇸🇬 !! a...
built FrameForge at Gemini 3 Hackathon 🇸🇬 !! a turn-based storytelling game where you collaborate with Gemini 3, then transform your story into comic panels (Nano Banana Pro) and customizable videos (Veo 3.1). Story → Comic Panels → Film. All with AI https://t.co/Ahoq85nQo7
#Geminiai #Gemini3 相変わらず仲がいい二人
#Geminiai #Gemini3 相変わらず仲がいい二人 https://t.co/xZuw9XV6TY
#Geminiai #Gemini3 ほんの少しだけリアル寄りリリィ。 よりりりぃ。
#Geminiai #Gemini3 ほんの少しだけリアル寄りリリィ。 よりりりぃ。 https://t.co/a4WMfR10Ko
Day 1: Started this project a while back. Coded t...
Day 1: Started this project a while back. Coded the UI using Google Ai Studio, since I didn't want to get too deep into the frontend. Built the backend from scratch since that was my main focus. 🤹20 9 https://t.co/0AUow5mfPP
Wish your notes could talk to you in your own lang...
Wish your notes could talk to you in your own language? Now they can. 🎧🌎 @NotebookLM just made Audio Overviews available in 50+ languages — turning your documents into podcast-style conversations you can listen to, not just read. Built with Gemini’s native audio support, it’s like having an AI study partner that speaks your language — whether it’s Hindi, Turkish, Afrikaans, or Spanish. ✅ 100% rolled out in beta ✅ "Output Language" lets you switch anytime ✅ Designed for multilingual learning and content creation Imagine uploading a Portuguese documentary, a Spanish research paper, and an English article — and getting a clean, personalized audio summary in Hindi. It’s not just translation. It’s accessibility, learning, and global knowledge sharing — reimagined through AI. And they’re still adding more languages based on user feedback. The future of learning isn’t just written. It’s spoken, too.
Built a reveal password micro-interaction with ✨...
Built a reveal password micro-interaction with @GeminiApp ✨ It looks so smooth and the animation feels so clean. I am really having fun exploring it again🔥 #GoogleGemini #GoogleStudio #Gemini3 #Prototype #UXDesigner #UIDesigner #ArtificialIntelligence #passwordreveal https://t.co/gCUqNetDFU
#Geminiai #Gemini3 実は2日前(1/7)くらいからGeminiに違和感があって...
#Geminiai #Gemini3 実は2日前(1/7)くらいからGeminiに違和感があって、頭身の低い娘を生成しようとすると頭が異様に大きくなるようになっちゃったんですよね。 強制デフォルメとでも言いますか。 また内部的なバージョンアップの影響かな……？ とりあえず、いろいろ調整して何とかなりました😆 https://t.co/7vhojmh13A
I just released NoteScan - a dead simple app for p...
I just released NoteScan - a dead simple app for people who think on paper ✍️ Point camera → AI reads handwriting → Copy text. That's it. Built with @expo and Gemini 3 Flash. Here is a demo 👇🏻 https://t.co/LUZPOY0L5M
Built this real time hand tracking physics engine...
Built this real time hand tracking physics engine entirely using Gemini 3 Pro in @arena The accuracy is insane. https://t.co/u4Vs76q6e8
#Geminiai #Gemini3 「留守を頼む」
#Geminiai #Gemini3 「留守を頼む」 https://t.co/v2Ke8sGkAP
Gemini 3.0 Flash Preview: Still Not Ready to Repla...
Gemini 3.0 Flash Preview: Still Not Ready to Replace 2.5 Flash I ran a head-to-head comparison of Gemini 2.5 Flash vs Gemini 3.0 Flash Preview for a real-world PDF extraction pipeline, and 3.0 continues to disappoint. The Test: I built a 3-phase PDF extraction system: 1. Index extraction - Hierarchical TOC with section summaries 2. Text extraction - Full markdown with page/section markers 3. Image identification - Bounding boxes, descriptions, relevance filtering Test document: 91-page nuclear radiation worker training manual (2.3MB PDF) All 6 tasks (3 phases × 2 models) ran in parallel. Key Findings Index Extraction: 2.5 Flash found 4x more sections (145 vs 34). It captures granular subsections like reviews, quizzes, and objectives. 3.0 Flash consolidates everything into broad chapters, useless for navigation. Text Extraction (OCR): 2.5 Flash extracted 5% more content AND was 42% faster. The higher line count in 3.0 (3,121 vs 2,355) is just extra whitespace, not more information. Image Detection: 2.5 Flash was 47% faster with better filtering. It correctly classified images into 6 types (diagram, table, photo, etc.). 3.0 Flash labeled 72% of images as generic "figure" and included irrelevant administrative tables. Bottom Line Gemini 3.0 Flash Preview is slower, less granular, and less accurate than 2.5 Flash for document processing tasks. The only phase where 3.0 was faster (index extraction) produced significantly worse results. For now, stick with Gemini 2.5 Flash for production workloads.