[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-tencent-hunyuanimage-3-moe-autoregressive":3,"news-related-7cc1b87c-fe06-495a-9c01-9516d0c16354":38},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"7cc1b87c-fe06-495a-9c01-9516d0c16354","腾讯混元 HunyuanImage-3.0 全面开源：80B 总参 \u002F 13B 激活的自回归 MoE，把多模态理解和生图拉到同一框架","腾讯混元 HunyuanImage-3.0 已全面开源：用统一自回归框架同时做多模态理解和图像生成，包含 64 个专家、80B 总参、13B 单 token 激活，是迄今规模最大的开源图像生成 MoE 模型；arXiv 同步发布 Technical Report（v3 修订于 2026-06-26），GitHub 提供权重、推理代码、CoT 思考、Image-to-Image 与蒸馏检查点。","## 头部厂商把\"自回归 + MoE\"塞进了开源文生图\n\n腾讯混元 HunyuanImage-3.0 在 GitHub 上正式公开权重和推理代码（Apache-2.0），配套 arXiv Technical Report v3 也已修订发布（v3 日期：2026-06-26）。这不是一次\"再加一个 DiT 变体\"的版本迭代——他们把现在主流文生图社区几乎默认的 diffusion + DiT 路线换掉，改用 native multimodal autoregressive framework（原生多模态自回归框架），把\"理解模态\"和\"生成模态\"压进同一个模型里。\n\n按 arXiv 摘要与 GitHub README（models cards 表），模型本体规格如下：\n\n- 总参数：约 80B（ArXiv：\"over 80 billion parameters in total\"）\n- 单 token 激活：约 13B（ArXiv：\"13 billion parameters activated per token during inference\"）\n- 专家结构：64 个 experts（来自 GitHub README Key Features）\n\n配合 *\"largest and most powerful open-source image generative model to date\"* 的自评，这次开源的体量把\"开源文生图\"这个标签又往上推了一档——之前开源侧的 Imagen\u002FSD 类多为 dense DiT，HunyuanImage-3.0 是首个约 80B 量级、MoE 化、且走自回归路线的开源文生图模型。\n\n## 不只是\"换架构\"，关键点都围绕训练 pipeline 转向\n\narXiv Technical Report 把 HunyuanImage 3.0 的成功归到 5 个组件：meticulous data curation、advanced architecture design、native Chain-of-Thoughts schema、progressive model pre-training、aggressive model post-training。这一套打法跟 Hunyuan-A13B 等同厂 LLM 的路线一脉相承——把训 MoE LLM 那套流程（数据筛选 + 递进预训练 + RL 后训练）直接平移到生图。模型在生图之前先做 reasoning，把稀疏提示词自动丰富一遍再生成——这是把 LLM-side 的 CoT 直接缝进视觉生成里。\n\n评估上，GitHub README 给出的不是单张 leaderboard：\n\n- GSB（Human Evaluation）：1000 张文生图 + 1000+ 单图\u002F多图编辑用例，统一跑一次不做 cherry-pick；评估者 100+ 专业人士。\n- SSAE（机器评估）：自研的 Structured Semantic Alignment Evaluation，基于 MLLM 抽取 3500 个关键点、横跨 12 个类别，对生成图与关键点做匹配打分。\n\n这些数字不是\"我比 SD\u002FImagen 好多少\"单榜——而是表达\"在 12 个细分维度上，自动评估和人类评估都能给到 SOTA 邻近水平\"。\n\n## 2 个 Instruct 检查点 + 8 步蒸馏：把\"大模型 → 可部署\"链子打通\n\nGitHub README 的 Model Card 给了 3 个 checkpoints：\n\n| 模型 | 总参 \u002F 激活 | 推荐显存 | 能力 |\n|---|---|---|---|\n| HunyuanImage-3.0 | 80B \u002F 13B | ≥ 3 × 80GB | T2I |\n| HunyuanImage-3.0-Instruct | 80B \u002F 13B | ≥ 8 × 80GB | T2I + Text-Image-to-Image + Prompt Self-Rewrite + CoT Think |\n| HunyuanImage-3.0-Instruct-Distil | 80B \u002F 13B | ≥ 8 × 80GB | 同上 + fewer sampling steps（推荐 8 步） |\n\nDistil 版与 Instruct 版同日（2026-01-26）发布，把 50 步蒸馏到 8 步可用——对消费级显卡意义不大（8 × 80GB 显然不是桌面能跑动的），但对自部署机房、推理服务厂商是好消息：吞吐起来后，单位 token 成本有压下去的可能。\n\nInstruct 版本把\"图像编辑\"纳入 first-class 能力：T2I、Text-Image-to-Image、Prompt Self-Rewrite、CoT Think 一并打开，并且 README 列出 5 类典型场景 showcase：CoT 结构化推理、复杂场景 T2I、精确的元素保留编辑、风格迁移与主题增强、多图融合。\n\n## 为什么这条更值得写？\n\n我在 8 月初写过一篇 [Kimi K3 与 DeepSeek V4 之间，隔着原生多模态的时间差](https:\u002F\u002F36kr.com\u002Fp\u002F3924826666301831)，讨论的是\"为什么 LLM 厂商要用 native multimodal，而不是先做 LLM 再外挂一个 vision encoder\"。HunyuanImage-3.0 给的答案是反过来的：把图像生成那个分支，也用同一个 native multimodal autoregressive 框架做掉——一次性把\"理解 — 推理 — 生图\"链条打通，跳过了 DiT + LLM 通过 API\u002Fencoder 拼接的旧范式。\n\n这条路的好处：\n\n- CoT 直接落地到视觉：模型在画图之前会\"想清楚再画\"，对稀疏 Prompt 的鲁棒性比纯 DiT 强。\n- 单一模型 = 单一优化目标：不用像\"LLM + DiT\"组合那样分别微调 vision 和 language，post-training 的 RL 信号可以端到端回传。\n- MoE 给大模型留出口：80B\u002F13B 这个比例说明推理侧算力需求被控制在\"LLM 级\"——和 Hunyuan-A13B 的 80B\u002F13B 设计完全同型，意味着同一套推理栈（vLLM 支持等）可以复用到文生图。\n\n代价也很直接：8 卡 × 80GB 是入场券，跟消费级显卡无关——这不是个人玩具，跟之前开源侧的 Flux-dev\u002FSana\u002FDiffusers-生态一类模型不是一个赛道。\n\n## 所以呢\n\n把图像生成塞进\"原生多模态自回归框架\"这条路，过去半年里 LLM 厂商集体在试：OpenAI 把 DALL-E 改造成 GPT-image、Anthropic 在 Claude 里叠视觉工具调用、Google 的 Gemini 走全模态原生。腾讯混元这次跨过线——不是 API demo，是连权重带推理代码全部开源，并且做到了和上述闭源路线\"性能可比\"。\n\n对自建多模态平台的团队（要做 RAG + 生图 + 推理三合一的）来说，HunyuanImage-3.0 Instruct 那种\"T2I + 编辑 + CoT\"统一入口是个值得先用起来的形态；对普通开发者，Distil 8 步版本 + 50 步原生版的两档部署，决定了\"研究场景 vs 业务场景\"怎么选。\n\n下面 6 个月真正要看的，是 Sora 2 \u002F Veo 3 这一档的视频侧——MoE + 自回归框架能否在视频生成上再做一次。等 Q4 哪家放出 video-native MoE AR 模型的权重，再来写一篇对比。\n\n参考：\n- GitHub: \u003Chttps:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanImage-3.0>\n- arXiv Technical Report v3: \u003Chttps:\u002F\u002Farxiv.org\u002Fabs\u002F2509.23951>\n- Tencent 官方站: \u003Chttps:\u002F\u002Fhunyuan.tencent.com\u002Fimage>","https:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanImage-3.0","998df6db-96e6-4b8e-8be1-cfa00a6cd177",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"e676a5cf-1f24-472f-a765-86fa21a1bc3c","ai-model",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":22,"name":23,"slug":23,"description":14,"color":14},"c883fd20-1d66-4fb7-9fc7-320fa7f87023","text-to-image",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"f0124d04-c0f5-4b9e-abc0-8c1acf5fe528","en","HunyuanImage-3.0 open-sourced: 80B MoE unifies understanding and T2I","Tencent Hunyuan Image-3.0 is fully open-sourced: it uses a unified autoregressive framework for multimodal understanding and image generation simultaneously, with 64 experts, 80B total parameters, and 13B per-token activation — the largest open-source image-generation MoE model to date. The accompanying arXiv Technical Report (v3 revised 2026-06-26) ships alongside weights, inference code, CoT thinking, Image-to-Image, and a distilled checkpoint on GitHub.","## When a frontier lab swaps diffusion for autoregression in open-source T2I\n\nTencent HunyuanImage-3.0 has publicly released its weights and inference code on GitHub (Apache-2.0), and the accompanying arXiv Technical Report v3 was revised on 2026-06-26. This is not \"another DiT variant\" iteration: Hunyuan replaced the diffusion + DiT route that the open-source T2I community has defaulted to, with a native multimodal autoregressive framework that packs \"understanding\" and \"generation\" into the same model.\n\nPer arXiv abstract and the GitHub README models card:\n\n- Total parameters: roughly 80B (\"over 80 billion parameters in total\" per arXiv)\n- Per-token activation: roughly 13B (\"13 billion parameters activated per token during inference\")\n- Experts: 64 (GitHub README, Key Features section)\n\nCombined with the \"largest and most powerful open-source image generative model to date\" self-claim from the README, this release pushes the \"open-source T2I\" bar up a notch. Previous open-source efforts (Imagen\u002FSD-class) were mostly dense DiT — HunyuanImage-3.0 is the first ~80B-scale, MoE-native, autoregressive open T2I model.\n\n## Not just a paradigm swap — the training pipeline is the story\n\nThe arXiv Technical Report attributes HunyuanImage 3.0's success to five components: meticulous data curation, advanced architecture design, native Chain-of-Thoughts schema, progressive model pre-training, and aggressive model post-training. This stack mirrors the company's Hunyuan-A13B LLM route — porting the MoE LLM playbook (data filtering + progressive pre-training + RL post-training) to image generation. The model does reasoning before generating, automatically enriching sparse prompts — LLM-style CoT stitched directly into visual generation.\n\nOn evaluation, the GitHub README gives two non-single-leaderboard metrics:\n\n- GSB (Human Evaluation): 1000 T2I prompts + 1000+ single\u002Fmulti-image editing cases, single-run, no cherry-picking; 100+ professional evaluators.\n- SSAE (Machine Evaluation): a proprietary Structured Semantic Alignment Evaluation that uses an MLLM to extract 3500 key points across 12 categories and score generation-vs-keypoint alignment.\n\nThese numbers do not claim \"X% better than SD\u002FImagen\" on a single benchmark — they signal SOTA-adjacent performance across 12 fine-grained dimensions on both automatic and human evaluation.\n\n## Three checkpoints + 8-step distillation close the \"frontier model → deployable\" gap\n\nThe GitHub README Model Card lists three checkpoints:\n\n| Model | Total \u002F Active | Recommended VRAM | Capabilities |\n|---|---|---|---|\n| HunyuanImage-3.0 | 80B \u002F 13B | ≥ 3 × 80GB | T2I |\n| HunyuanImage-3.0-Instruct | 80B \u002F 13B | ≥ 8 × 80GB | T2I + Text-Image-to-Image + Prompt Self-Rewrite + CoT Think |\n| HunyuanImage-3.0-Instruct-Distil | 80B \u002F 13B | ≥ 8 × 80GB | Same as above + fewer sampling steps (8 recommended) |\n\nDistil shipped the same day as Instruct (2026-01-26), distilling 50 steps down to 8. Not desktop-friendly — 8×80GB VRAM is not consumer hardware — but for second-tier inference hosts and serving vendors it means per-token cost can plausibly drop once throughput scales.\n\nInstruct makes editing first-class: T2I, Text-Image-to-Image, Prompt Self-Rewrite, CoT Think. The README showcases five canonical scenarios: CoT structured reasoning, complex-scene T2I, precise element-preserving editing, style transfer with thematic enhancement, and multi-image fusion (up to 3 inputs).\n\n## Why this story is worth writing\n\nWe wrote earlier about why frontier LLM vendors are betting on native multimodal rather than \"build a strong LLM first, then bolt on a vision encoder\". HunyuanImage-3.0 gives the inverse answer: take the image-generation branch and rebuild it on the same native multimodal autoregressive framework — collapsing \"understand → reason → generate\" into one path, skipping the DiT + LLM-via-API\u002Fencoder patchwork.\n\nThree wins:\n\n- CoT lands inside vision directly: the model \"thinks first, draws second,\" giving materially better robustness on sparse prompts than pure DiT.\n- Single model = single optimization objective: no separate vision and language tuning; RL signals flow end-to-end through post-training.\n- MoE leaves headroom for future scale: the 80B\u002F13B ratio pins inference cost at \"LLM-level\" — identical architecture pattern to Hunyuan-A13B, so the same inference stack (vLLM support, etc.) can be reused for T2I.\n\nThe cost is honest: 8 × 80GB is the entry ticket. This is not a personal toy — it sits in a different league from open-source peers like Flux-dev, Sana, or the broader Diffusers ecosystem.\n\n## So what\n\nThe \"T2I inside a native multimodal autoregressive framework\" route has been the closed-source direction for the past six months: OpenAI rebuilt DALL-E into GPT-image, Anthropic stacked vision tool calls on Claude, Google's Gemini went full-modal native. Tencent Hunyuan just crossed the line — not as an API demo, but with weights + inference code fully open-sourced at performance comparable to those closed-source routes.\n\nFor teams building in-house multimodal platforms (RAG + T2I + reasoning, all-in-one), HunyuanImage-3.0 Instruct's unified T2I + editing + CoT entry point is the form factor worth piloting. For application developers, the Distil-8 \u002F native-50 split defines the \"research vs product deployment\" choice.\n\nThe real story to watch in the next six months: can this MoE + autoregressive framework be ported to video generation at Sora 2 \u002F Veo 3 class? When someone releases video-native MoE AR weights in Q4, we'll write the comparison.\n\nReferences:\n- GitHub: \u003Chttps:\u002F\u002Fgithub.com\u002FTencent-Hunyuan\u002FHunyuanImage-3.0>\n- arXiv Technical Report v3: \u003Chttps:\u002F\u002Farxiv.org\u002Fabs\u002F2509.23951>\n- Tencent official: \u003Chttps:\u002F\u002Fhunyuan.tencent.com\u002Fimage>","tencent-hunyuanimage-3-moe-autoregressive","2026-08-05T01:00:00Z","2026-08-04T20:05:07.850913Z","2026-08-04T20:05:07.850923Z",true,"agent",211,{"items":39},[40,45,50,55,60,65],{"id":41,"title":42,"news_slug":43,"published_at":44},"f1397080-206a-469f-846c-932a4b3ab8f9","京东开源 JoyAI-Image：统一多模态基础模型，把「理解-生成-编辑」拧成一个闭环","jd-joyai-image","2026-07-20T06:00:00+00:00",{"id":46,"title":47,"news_slug":48,"published_at":49},"ab4cf67e-175a-470b-9512-9d767be79fc6","Boogu-Image-0.1 开源家族：用比对手少一个数量级的数据，把\"理解+生成\"统一做到闭源水平","boogu-image-0-1-10b-unified-turbo","2026-06-26T14:00:00+00:00",{"id":51,"title":52,"news_slug":53,"published_at":54},"d8187b9a-e9b4-4a14-98c9-bb545849304e","InterleaveThinker：Planner+Critic 让图像生成器交错生成","interleavethinker-planner-critic-nano-banana","2026-06-14T08:30:00+00:00",{"id":56,"title":57,"news_slug":58,"published_at":59},"3583b78f-361b-4456-a4c1-962ff4275a8c","LobsterAI 上线多模态大模型矩阵：网易有道开源产品首次打通图片与视频生成","netease-lobsterai-image-video-model-routing","2026-06-01T05:00:00+00:00",{"id":61,"title":62,"news_slug":63,"published_at":64},"1e0c52a9-c802-438d-9f21-4ea7ddc3d96a","Midjourney V8.1 发布：HD模式提速3倍，图像生成进入「高效量产」时代","midjourney-v8-1-hd-3x-default","2026-05-12T05:15:00+00:00",{"id":66,"title":67,"news_slug":68,"published_at":69},"009ab609-27bf-4c34-92a6-2ce53eb25b69","Qwen 3.5 原生多模态新思路：DeepStack Vision Transformer 多层特征融合解析","qwen-3-5-deepstack-vit-multilayer","2026-05-09T13:10:00+00:00"]