ESP32 恐龙玩偶：Jev 决定何时给侄子看图
用 ESP32-S3 加 Gemini 语音和 FLUX 生图做意大利语恐龙玩偶，Jev 1.13 负责判断对话中该不该出图并审核画面是否适合小孩。
my 3 yr old nephew is in love with dinosaurs, so i built him a little dinosaur friend he can talk to 🦖 using an ESP32 device + gemini llm/voice + jev for decisions + flux for images. my nephew speaks italian and he's curious about everything, and he's starting to learn English too, so i gave Dino context about him: his family, his cat, the food he likes, kindergarten, his favourite things. little details that help make the conversation feel familiar. the idea is to learn through play. talk about dinosaurs, ask little questions, count things together, introduce an English word when it fits. for example, a conversation about his cat can show the word "CAT" and a cat drawing on the screen. a dinosaur story can become a little counting game. or he can ask Dino to draw a velociraptor on a scooter 😄 it keeps track of recent exchanges during the session, so he can ask follow-up questions without starting over. JEV decides when to show images during the conversation, helping him to stay engaged and also curates the content that is being shown (to keep it the content safe for him) the hardware is a @waveshare00 ESP32-S3 Touch AMOLED 1.8, with a microphone, speaker and wifi. press a button, speak, and Dino talks back, with expressions, words and pictures on the display. built with Astra. the other AI models are connected through @OpenRouter: - Gemini Flash for transcriptions and handling the conversation - Gemini 3.1 Flash TTS for Dino's voice - FLUX.2 Klein 4B for drawings - JEV 1.13 for deciding whether a picture would help and checking the proposed scene's suitability still a prototype, but this is what i find beautiful about where we are now. get a $25 tiny device, these models, and something a kid loves, and build a little companion personalized just for him.