#open-source
14 articles
所有標籤 gemini-3-5-flash google google-io ai-models agent-tools deepmind gemini-3-5 gemini-3-5-pro gemini-omni gemini-omni-flash video-generation ai-video gemini-4 claude-mythos gpt-5-5 seedance bytedance pricing model-release gpt-5-6 openai chatgpt alignment leak gemini omni veo hidream open-source image-generation diffusion-transformer multimodal tutorial rumor llm api comparison guide best alternative openrouter aws azure google-cloud inference image-to-3d 3d 3d-generation tripo3d h3.1 pbr quad-mesh announcement wavespeedai multiview-to-3d text-to-3d text-to-image nucleus nucleus-image image-to-image materials texture patina image-to-map game-dev unreal unity blender material-extract video-to-video video-inpainting object-removal masking void sam3 text-to-audio music-generation music-cover style-transfer minimax image-to-video pixverse animation transition reference-to-video character-consistency text-to-video baidu ernie multilingual chinese fast audio-to-video music-video lip-sync runway-ml ideogram image-editing sora portrait-effect photo-styling parkour action-video talking-photo travel-photo ad-generation spokesperson virtual-try-on fashion ai-avatar free-tools avatar-generator talking-head ai-image image-generator video-generator wan-2-7 kling gpt-image-2 predictions deevid-ai wan alibaba video-editing video-edit wan-2.7 video-extend veo3 start-end-to-video kuaishou elements vace video-joiner wavespeed-ai gemma-4 on-device-ai audio-converter audio-processing file-conversion image-converter image-processing face-blur privacy video-converter video-processing ai-tools 4k midjourney flux nano-banana seedream best-ai-image-generator pixverse-v6 audio video-effects glm zhipu-ai claude gpt deepseek ai-news phota image-enhance upscaler image-quality photorealistic camera-control vfx anthropic cybersecurity ai-music suno lyria magihuman davinci sand-ai digital-human audio-video davinci-magihuman professional ai-image-generator qwen-image pollo-ai lovart freepik ai-video-generator vidu best-ai-video-generator higgsfield kling-image-o3 ai-image-generation girl-filter face-transformation portrait smile-filter photo-editing watermark-removal sora-alternative sora-shutdown pika grok ltx veo-4 photo-colorizer colorize photo-restoration body-swap face-swap portrait-transfer prismaudio video-to-audio foley ai-audio sound-generation hunyuan audio-generation v2a iclr recraft recraft-v4 text-to-vector svg design dall-e vocal-remover karaoke music-production stem-separation people-remover inpainting fotor photo-editor content-creation desktop-app mp3 wav flac aac png jpg webp heic mp4 mov avi webm janitor-ai media-io video-editor m2.7 ai-model agent coding benchmark age-filter entertainment aging dog-selfie pet-content gender-swap ghibli-filter anime studio-ghibli midjourney-v8 stable-diffusion best-tools ai-content-detector content-moderation content-safety nsfw-detection text-moderation image-moderation video-moderation moderation-api developer-guide sora-2 sketch-to-video infinitetalk celebrity-look-alike face-recognition clothes-changer fat-filter meme fortune-teller math-solver education story-generator creative-writing review baseten 2026 canva fal-ai fireworks-ai leonardo-ai modal gpu-cloud replicate cloudflare runpod together-ai ai-research helios bitdance bitdance-14b autoregressive qwen-image-2 typography skyreels skyreels-v3 talking-avatar portrait-animation soulx flashhead soulx-flashhead real-time streaming nano-banana-2 nano-banana-pro ai-images q3 wavespeed-desktop android mobile playground batch-processing lora workflow ai-pipeline ffmpeg audio-conversion image-conversion video-conversion video-merge video-trimming video-enhancement video-upscale inworld tts text-to-speech voice-ai coming-soon gpt-image kimi moonshot-ai ai-assistant local-ai personal-ai prediction genie-3 world-model interactive-environments mova clawdbot personal-assistant automation chatbot javascript typescript sdk python speculation ai-collaboration productivity ai-agents no-code app-builder development apple background-remover face-enhancer image-enhancement image-eraser inpaint tools claude-code codex ai-coding cursor developer-tools image-enhancer ai-platforms hedra avatars heygen creative video-marketing ideas adobe firefly quality rankings image-translation localization image-upscaling enhancement video-extension video-upscaling enterprise developer clipdrop stability-ai dalle deepai performance black-forest-labs vertex-ai infrastructure hailuo-ai hugging-face tencent text-rendering imagen kling-ai serverless luma-ai dream-machine nightcafe ai-art pika-labs lm-arena runway digital-twins tips video-production synthesia dalle-3 prompting avatar multi-modal aimlapi byteplus comfyui dreamina kie-ai openart poyo-ai topaz upscaling skywork qwen training fine-tuning depth controlnet pose upscale outpaint canny lightricks sdxl background-removal marketing event e-commerce product-photography mochi cogvideo social-media instagram
HiDream-O1-Image-Dev：以80億參數擊敗560億FLUX.2的原生像素模型
HiDream-O1-Image-Dev是一個80億參數的蒸餾圖像模型，捨棄了VAE與外部文字編碼器，原生生成2K解析度，並在GenEval、DPG和HPSv3基準上超越了7倍體積的模型。
3 分鐘閱讀
什麼是Google Gemma 4？架構、基準測試及其重要性
Google Gemma 4是DeepMind迄今最強大的開源模型系列，以Apache 2.0授權提供四種規格，支援多模態輸入、原生推理，並可部署至樹莓派等端側設備。
4 分鐘閱讀
daVinci-MagiHuman：碾壓所有數位人生成器的開源模型
daVinci-MagiHuman 是一個 150 億參數的開源模型，能在單張 H100 上以 2 秒生成口型同步的說話人頭影片。勝率優於 Ovi 1.1（勝率 80%）與 LTX 2.3（勝率 60.9%），採用 Apache 2.0 授權，支援多語言，速度極快。
2 分鐘閱讀
daVinci MagiHuman Text-to-Video 現已登陸WaveSpeedAI
daVinci MagiHuman Text-to-Video 可從文字提示生成電影感十足的人物中心影片，支援可選音訊同步。150億參數開源模型，最高 1080p 畫質，影片長度 5-10 秒。提供 REST API，每秒 $0.04，無冷啟動延遲。
2 分鐘閱讀
daVinci MagiHuman Image-to-Video 現已登陸WaveSpeedAI
daVinci MagiHuman Image-to-Video 是一個 150 億參數的開源模型，可將參考圖片動畫化為電影級影片，並支援可選的音頻同步。效果媲美 WAN 2.5。解析度最高 1080p，時長 5-10 秒。REST API，$0.04/秒，無冷啟動。
1 分鐘閱讀
Helios：一個不走捷徑的即時長影片生成模型
Helios 能在單張 H100 上以 19.5 FPS 生成長達一分鐘的影片——不使用 KV 快取、稀疏注意力機制或任何常見的加速技巧。以下是它與眾不同之處。
2 分鐘閱讀
BitDance 14B：比其他自迴歸模型快30倍的AI圖像生成
BitDance 14B 使用二進位標記生成圖像，速度比其他自迴歸模型快30倍，並在基準測試中超越 FLUX.1。立即在WaveSpeedAI上體驗。
2 分鐘閱讀
Kimi K2.5：關於月之暗面視覺智能體模型的一切
Kimi K2.5 是月之暗面的開源 1T 參數模型，具備智能體群集技術、256K 上下文和多模態功能。以下是完整解析。
3 分鐘閱讀
OpenClaw：您可以控制的開源個人AI助手
探索 OpenClaw，一款創新的開源個人AI助手，在您自己的設備上運行，與多個消息平台集成，同時完全由您掌控。
2 分鐘閱讀
MOVA vs WAN vs Sora 2 vs Seedance：2026年影片音訊AI模型比較
比較OpenMOSS MOVA、WAN 2.2 Spicy、WAN 2.6 Flash、Sora 2和Seedance 1.5 Pro的影片生成與音訊功能。特性、定價和建議。
4 分鐘閱讀
DeepSeek V4 現已登陸WaveSpeedAI：關於這款即將推出的編碼 AI 模型的一切
DeepSeek V4 將於 2026 年 2 月推出，具有革命性的編碼能力。以下是我們對其架構、功能和基準的了解。
1 分鐘閱讀
Apple SHARP: 一秒內將任何照片轉換成3D
Apple的SHARP AI模型使用高斯斑點技術，在一秒內將單張2D照片轉換成逼真的3D場景。了解這項開源突破的運作原理。
2 分鐘閱讀