[{"_1":2,"_255":-5,"_256":-5},"loaderData",{"_3":4,"_7":8},"root",{"_5":6},"footerPartners",[],"routes/tools/index",{"_9":10,"_67":57,"_68":57,"_69":57,"_70":71,"_72":39,"_28":73,"_254":33},"tools",[11],{"_12":13,"_14":15,"_16":17,"_18":19,"_20":21,"_22":23,"_24":25,"_26":27,"_28":29,"_30":31,"_32":33,"_34":35,"_36":17,"_37":35,"_38":39,"_40":41,"_42":43,"_45":46,"_47":48,"_49":50,"_51":48,"_52":53,"_54":55,"_56":57,"_58":48,"_59":-5,"_60":48,"_61":62,"_63":64,"_65":66},"id",4312,"code","qwen35-4b-3d-coder","name","Qwen3.5-4B-3D-Coder","icon","/uploads/agent/agent-1793978831011.webp","screenshot","/uploads/agent/agent-1793978832995.png","description","Qwen3.5-4B-3D-Coder 是 CCORE 独立微调的 Three.js 编程模型，输出可编辑 HTML/JavaScript 场景，提供 GGUF 与 FP16 权重。","detail","

资料核查于 2026 年 11 月 6 日。正文区分发布方声明与第三方体验，不代表 AIDreamHub 已完成产品实测。

真正交付的是场景代码

Qwen3.5-4B-3D-Coder 是 CCORE 基于 Qwen3.5-4B 的独立微调，并非 Qwen 官方版本。它主要输出 Three.js 的 HTML/JavaScript，适合程序化环境、教学演示和浏览器原型，与根据照片生成网格不是同一任务。

画面可信也可能有重复变量、坏掉的相机控制或昂贵几何。只有需要可编辑代码、且能实际运行和修错时，才适合选择它。目标若是带骨骼角色或生产 GLB，先确认另一条资产流程。

AIDreamHub 截取的官方项目页面，不是本地运行画面。

检查权重版本与生成设置

页面提供 Q4_K_M、Q8_0 GGUF 和 FP16 Safetensors，并标注 10 月 27 日检查点更新。应记录具体修订号，不把同名下载视为同一版本。公开数据集含 4,146 个例子，拆为 3,933 训练与 213 验证；这不是任意生产场景的成功率评测。

发布方为所列量化建议温度 0.6–0.8、top-p 0.90–0.95、top-k 20–40、重复惩罚 1.0，并要求 LM Studio 关闭 MTP。它们是可记录的起点，不保证其他运行时默认值等价。

\"发布方的 — 发布方的 Three.js 渲染示例，不代表独立测得的成功率。

How to Use：先渲染一个受约束场景

选择兼容的本地运行器与明确 GGUF 包，使用下方官方 Three.js 核心模块约束。模型卡提供 Ollama 路线。先核查运行器的硬件和安装要求，不要把无关的一键脚本直接放进高权限 shell。

第一次只要一个清晰视角、少量物体和一种交互。保存 HTML，通过本地 HTTP 服务打开并检查控制台。保存成功还不算渲染成功；本站没有运行模型推理。

ollama run hf.co/ccore/Qwen3.5-4B-3D-Coder:Q4_K_M\n\nThree.js core module:\nhttps://cdnjs.cloudflare.com/ajax/libs/three.js/0.160.0/three.module.min.js\nNo addons or legacy examples/js imports.\n\nLocal preview, from the folder containing the saved HTML:\npython -m http.server 8000 --bind 127.0.0.1

示例与社区材料能说明多少

官方展示建筑、花园与体素场景，并承认有些生成仍需重试或修正，已观察到重复变量等问题。这说明预期输出和已知缺陷，不建立可靠的一次成功率。

第三方运行指南区分下载体积与内存，也区分程序场景与网格，这是有用的解释，但不等于受控质量基准。短视频教程和精选画廊不能扩展为普遍可靠性结论。

渲染—检查—修复的闭环

固定任务、版本和设置，先否决语法错误、导入失败和空画布，再检查取景、控制、缩放适配、光照与性能。保留所有尝试和人工修补记录，不让最终截图掩盖实际工作量。

在可丢弃浏览器环境运行不可信 JavaScript，不放进带凭据的生产页面。限制网络依赖。需要导出时，在目标引擎检查结果；不能把另一款 GLB-3D-Coder 预览的能力并入本模型。

关卡	检查	意义
执行	控制台、导入、非空画布	文本完成不等于运行
交互	相机、缩放与目标动作	截图看不见控制失效
交付	浏览器与必要导出	HTML 不等于生产 GLB

\"编辑手绘：生成代码之后，还要实际渲染验收。不是产品输出或实测结果。\" — 编辑手绘：生成代码之后，还要实际渲染验收。不是产品输出或实测结果。

按输出类型比较替代品

本模型适合本地程序化场景草稿；复杂应用逻辑应比较通用编程模型；可复用几何应比较网格生成；实物忠实重建则需要多视图或测量证据。不要只凭名称都有 3D 就横向排名。

更有用的指标是单位编辑时间内通过验收的场景，而不是 token 速度。简单教学画面和带状态、数据、认证的应用，需要完全不同的验收深度。

路线	目标输出	优先检查
3D-Coder	可编辑 Three.js 页面	渲染与代码正确性
网格流程	可复用几何	拓扑、UV、引擎导入
通用编程模型	复杂交互应用	逻辑、测试和依赖

许可与合理预期

模型标注 Apache-2.0，发布的数据集标注 MIT；Three.js、字体、纹理等仍有各自条款。模型许可不授予参考图或他人设计的权利。

保持「独立作者」「预览」定位。手绘图解释代码需要真实渲染检查，不是产品截图或模型生成的三维效果。

采用前常见问题

会自动导出 GLB 吗？

主要文档针对 HTML/JavaScript 场景。发布方另有 GLB-3D-Coder 预览，不能混合能力。额外导出或网格转换需要接入，并在实际引擎中检查。

应该从什么设置开始？

按对应检查点说明使用温度 0.6–0.8、top-p 0.90–0.95、top-k 20–40、重复惩罚 1.0；LM Studio 还建议关闭 MTP。同时保存运行时和版本。

怎样确认场景真的能用？

通过本地 HTTP 服务打开 HTML，查看错误，测试相机、交互和缩放，并在目标设备尺寸与性能预算下检查。完整文本回复或好看的单张截图都不足以验收。

核查资料

继续寻找同类工具

","attributes",[],"categories",[],"paymentPlatform",[],"language","zh","url","https://huggingface.co/ccore/Qwen3.5-4B-3D-Coder","websiteName","websiteUrl","cate1Code","3d-modeling","cate2Code","ai-coding","tags",[17,44],"三维工具","views",2,"likes",0,"featured",false,"featuredSort","monthVisitedCount",["B","0"],"traffic",{},"status",1,"urlStatus","urlCheckedAt","urlFailCount","publishedAt",["D",1793775600000],"createdAt",["D",1793978847412],"updatedAt",["D",1793983558149],"total","page","totalPages","sort","latest","category",[74,85,91,99,107,115,123,130,137,144,152,159,165,172,179,185,191,198,205,212,219,226,233,240,247],{"_12":75,"_14":76,"_16":77,"_22":-5,"_32":33,"_78":57,"_79":80,"_81":82,"_63":83,"_65":84},198,"ai-agent","AI 智能体","level","parentCode","","toolCount",55,["D",1768983237537],["D",1768983237537],{"_12":86,"_14":41,"_16":87,"_22":-5,"_32":33,"_78":57,"_79":80,"_81":88,"_63":89,"_65":90},142,"AI 编程助手",53,["D",1768983180075],["D",1768983180075],{"_12":92,"_14":93,"_16":94,"_22":95,"_32":33,"_78":57,"_79":80,"_81":96,"_63":97,"_65":98},65,"seo","SEO优化","AI搜索引擎优化工具",21,["D",1768892383988],["D",1768892383988],{"_12":100,"_14":101,"_16":102,"_22":103,"_32":33,"_78":57,"_79":80,"_81":104,"_63":105,"_65":106},16,"video-generation","AI 视频生成","AI视频创作和编辑工具",20,["D",1768892377383],["D",1779004146835],{"_12":108,"_14":109,"_16":110,"_22":111,"_32":33,"_78":57,"_79":80,"_81":112,"_63":113,"_65":114},9,"image-generation","AI 图像生成","AI图像和艺术创作工具",19,["D",1768892376194],["D",1784549701580],{"_12":116,"_14":117,"_16":118,"_22":119,"_32":33,"_78":57,"_79":80,"_81":120,"_63":121,"_65":122},79,"productivity","生产力工具","AI生产力工具",18,["D",1768892385928],["D",1768892385928],{"_12":124,"_14":125,"_16":126,"_22":-5,"_32":33,"_78":57,"_79":80,"_81":127,"_63":128,"_65":129},170,"ai-music","AI 音乐",12,["D",1768983209261],["D",1768983209261],{"_12":131,"_14":132,"_16":133,"_22":-5,"_32":33,"_78":57,"_79":80,"_81":134,"_63":135,"_65":136},135,"ai-chat","AI 对话助手",11,["D",1768983173067],["D",1768983173067],{"_12":138,"_14":139,"_16":140,"_22":141,"_32":33,"_78":57,"_79":80,"_81":108,"_63":142,"_65":143},100,"text-to-speech","文字转语音","AI文字转语音工具",["D",1768892389112],["D",1768892389112],{"_12":145,"_14":146,"_16":147,"_22":148,"_32":33,"_78":57,"_79":80,"_81":149,"_63":150,"_65":151},93,"speech-recognition","语音识别","AI语音转文字工具",6,["D",1768892388178],["D",1768892388178],{"_12":153,"_14":154,"_16":155,"_22":-5,"_32":33,"_78":57,"_79":80,"_81":156,"_63":157,"_65":158},163,"ai-writing","AI 写作",5,["D",1768983201862],["D",1768983201862],{"_12":160,"_14":161,"_16":162,"_22":-5,"_32":33,"_78":57,"_79":80,"_81":156,"_63":163,"_65":164},191,"ai-translation","AI 翻译",["D",1768983230474],["D",1768983230474],{"_12":166,"_14":167,"_16":168,"_22":169,"_32":33,"_78":57,"_79":80,"_81":156,"_63":170,"_65":171},58,"marketing","营销工具","AI营销工具",["D",1768892383062],["D",1768892383062],{"_12":173,"_14":174,"_16":175,"_22":176,"_32":33,"_78":57,"_79":80,"_81":46,"_63":177,"_65":178},51,"design-tools","设计工具","AI图形设计工具",["D",1768892382134],["D",1768892382134],{"_12":180,"_14":39,"_16":181,"_22":182,"_32":33,"_78":57,"_79":80,"_81":57,"_63":183,"_65":184},121,"3D建模","AI 3D建模工具",["D",1768892391955],["D",1768892391955],{"_12":46,"_14":186,"_16":187,"_22":188,"_32":33,"_78":57,"_79":80,"_81":57,"_63":189,"_65":190},"text-generation","文本生成","AI文本内容生成工具",["D",1768892375210],["D",1768892375210],{"_12":192,"_14":193,"_16":194,"_22":195,"_32":33,"_78":57,"_79":80,"_81":48,"_63":196,"_65":197},30,"code-development","代码开发","AI编程和开发工具",["D",1768892379363],["D",1768892379363],{"_12":199,"_14":200,"_16":201,"_22":202,"_32":33,"_78":57,"_79":80,"_81":48,"_63":203,"_65":204},37,"writing-assistant","写作助手","AI写作和编辑工具",["D",1768892380266],["D",1768892380266],{"_12":206,"_14":207,"_16":208,"_22":209,"_32":33,"_78":57,"_79":80,"_81":48,"_63":210,"_65":211},107,"image-editing","图像编辑","AI图像编辑工具",["D",1768892390114],["D",1768892390114],{"_12":213,"_14":214,"_16":215,"_22":216,"_32":33,"_78":57,"_79":80,"_81":48,"_63":217,"_65":218},128,"education","教育学习","AI教育和学习工具",["D",1768892392869],["D",1768892392869],{"_12":220,"_14":221,"_16":222,"_22":223,"_32":33,"_78":57,"_79":80,"_81":48,"_63":224,"_65":225},72,"data-analysis","数据分析","AI数据分析工具",["D",1768892384920],["D",1768892384920],{"_12":227,"_14":228,"_16":229,"_22":230,"_32":33,"_78":57,"_79":80,"_81":48,"_63":231,"_65":232},114,"video-editing","视频编辑","AI视频编辑工具",["D",1768892391043],["D",1768892391043],{"_12":234,"_14":235,"_16":236,"_22":237,"_32":33,"_78":57,"_79":80,"_81":48,"_63":238,"_65":239},86,"translation","翻译工具","AI翻译工具",["D",1768892387047],["D",1768892387047],{"_12":241,"_14":242,"_16":243,"_22":244,"_32":33,"_78":57,"_79":80,"_81":48,"_63":245,"_65":246},44,"chatbot","聊天机器人","AI聊天机器人和对话AI",["D",1768892381169],["D",1768892381169],{"_12":248,"_14":249,"_16":250,"_22":251,"_32":33,"_78":57,"_79":80,"_81":48,"_63":252,"_65":253},23,"audio-generation","音频生成","AI音乐和音频创作工具",["D",1768892378444],["D",1768892378444],"locale","actionData","errors"]