blockbench-new-model

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Blockbench New Model

Blockbench 新模型

Build a new subject end to end: headless
.bbmodel
authoring, a Blender scene ready to render, and a
render.cmd
that renders PNG stills on every GPU and encodes them with FFmpeg.
Load blockbench-use before the first model edit and follow its routing to the domain skills. The form answers settle its appearance-versus-performance question; do not ask it again.
端到端构建新主体:进行无头
.bbmodel
创作,生成可直接渲染的 Blender 场景,并创建
render.cmd
,利用所有 GPU 渲染 PNG 静帧,再由 FFmpeg 对其进行编码。
首次模型编辑前加载 blockbench-use,并遵循其对各领域 skill 的路由。表单答案已经确定外观与性能方面的取舍;不要再次询问。

1. Intake form

1. 采集表单

Ask before any building. Use the client's structured question tool (
AskUserQuestion
in Claude Code); without one, send the same questions as one numbered list. Present options in the order shown, without marking a recommendation.
Round 1: what to build. Ask all four questions in one call.
HeaderQuestionOptions (label: description)
QualityWhat quality level?High-quality: detailed silhouette, smooth rounded parts, detail in textures · Low-poly: faceted, bold shapes, flat color regions
TexturesWhich texture style?PBR textured: albedo + normal + MER, resolution asked next · Vanilla Minecraft: Minecraft's own 16× textures · Classic Blockbench: hand-painted pixel art, no PBR
AnimationHow should it move?Animated: keyframed clips · Havok physics: physics-based motion baked with the Havok plugin in Blockbench desktop
SceneWhat kind of scene?Simple 3D scene: subject with a few context props · Complete 3D scene: full environment with foreground to background · Simple 3D vignette: subject on a small diorama base · 3D isometric room: cutaway room seen from an isometric camera
Round 2: details. Ask only the questions that apply, in one call.
HeaderWhenQuestionOptions
SubjectNo argument was givenWhat is the subject?Two or three example subjects suited to the round 1 answers; the user types their own through Other
ResolutionTextures = PBRWhich PBR resolution?2K PBR: 2048² maps, ~32 px per unit · HD PBR: 512–1024² maps, 128–256 px per block · PBR (standard): 16×–32× pixel art with PBR channels
OutputAlwaysWhere should the project folder go?A parent folder the user chose before, if you know one · Current directory:
./<slug>/
· anything else through Other
RenderAlwaysWhich render preset?1080p: 1920×1080, 30 fps, 10 s · 4K UHD: 3840×2160, 30 fps, 10 s · Square loop: 1080×1080, 30 fps, 6 s seamless loop · Quick preview: 1280×720, 24 fps, 5 s
An argument after the command (
/blockbench-new-model a lighthouse on a sea stack
) is the subject. Derive
<slug>
from it (lowercase, hyphens, ≤ 40 characters). Create
<output>/<slug>/
and refuse to overwrite a folder that already holds a project; offer a numbered slug instead.
Echo the full brief back in one sentence ("A high-quality, 2K PBR textured, animated simple 3D vignette of a lighthouse on a sea stack, 1080p 30 fps 10 s, in …") and continue without waiting for confirmation.
开始构建前必须先询问。使用客户端的结构化问题工具(Claude Code 中的
AskUserQuestion
);如果没有该工具,则将同样的问题作为一份编号列表发送。按所示顺序提供选项,不要标记推荐项。
第 1 轮:构建什么。 在一次调用中询问全部四个问题。
标题问题选项(标签:说明)
质量使用什么质量等级?高质量(High-quality):轮廓细致、部件圆润平滑、纹理细节丰富 · 低多边形(Low-poly):多面体外观、造型醒目、色彩区域扁平
纹理使用哪种纹理风格?PBR 贴图(PBR textured):albedo + normal + MER,下一步询问分辨率 · 原版 Minecraft(Vanilla Minecraft):Minecraft 自带的 16× 贴图 · 经典 Blockbench(Classic Blockbench):手绘像素艺术,不使用 PBR
动画主体应如何运动?动画(Animated):关键帧 clip · Havok 物理(Havok physics):在 Blockbench desktop 中使用 Havok plugin 烘焙基于物理的运动
场景需要哪种场景?简单 3D 场景(Simple 3D scene):主体搭配少量环境道具 · 完整 3D 场景(Complete 3D scene):包含从前景到背景的完整环境 · 简单 3D 小景(Simple 3D vignette):主体置于小型 diorama 基座上 · 3D 等距房间(3D isometric room):通过等距相机查看剖开的房间
第 2 轮:详细信息。 一次调用中只询问适用的项目。
标题何时询问问题选项
主体未提供参数主体是什么?提供两个或三个适合第 1 轮答案的示例主体;用户可通过
Other
输入自己的主体
分辨率纹理 = PBR选择哪种 PBR 分辨率?2K PBR:2048² maps,约每个单位 32 px · HD PBR:512–1024² maps,每个方块 128–256 px · PBR(standard):带 PBR 通道的 16×–32× 像素艺术
输出始终项目文件夹应放在哪里?如果已知,提供用户之前选择的父文件夹 · 当前目录:
./<slug>/
· 可通过
Other
输入其他位置
渲染始终使用哪个渲染预设?1080p:1920×1080,30 fps,10 s · 4K UHD:3840×2160,30 fps,10 s · 方形循环(Square loop):1080×1080,30 fps,6 s 无缝循环 · 快速预览(Quick preview):1280×720,24 fps,5 s
命令后的参数(
/blockbench-new-model a lighthouse on a sea stack
)即为主体。由该参数推导
<slug>
(小写、连字符、≤ 40 个字符)。创建
<output>/<slug>/
;如果目标文件夹中已有项目,则拒绝覆盖,并提供一个带编号的 slug 作为替代。
用一句话复述完整 brief(“一个高质量、采用 2K PBR 贴图、带动画的简单 3D 小景,主体是海蚀柱上的灯塔,1080p、30 fps、10 s,输出到……”),然后继续执行,无需等待确认。

2. Preflight

2. 预检

Check everything the answers need before modeling; report all gaps at once.
  • Headless server:
    bbmodel_*
    tools are connected. If not, see setup, give the user the command with their folder filled in, and stop.
  • Preview renders:
    bbmodel_render
    needs Node 23.6+ with npm and a GPU; its packages install on first use (rendering). Warn the user about the one-time install, and about
    $BB_RENDER_HOME
    if C: is short on space.
  • Blender, FFmpeg, GPUs: locate and probe them as in find the tools.
  • Havok physics: Blockbench desktop with the MCP plugin and the Havok Physics Animations plugin (
    havok_*
    tools after
    blockbench_launch
    ). If unavailable, ask whether to switch to keyframed animation.
  • AI textures (PBR):
    FAL_KEY
    for generated albedo. If missing, note it and paint procedurally.
  • glTF route: when Blender has no
    .bbmodel
    importer, Blockbench desktop is needed for glTF export.
检查答案所需的全部环境;一次性报告所有缺失项。
  • 无头 server(Headless server):
    bbmodel_*
    tools 已连接。如果未连接,请参阅 setup,向用户提供填入其文件夹的 command,然后停止。
  • 预览渲染(Preview renders):
    bbmodel_render
    需要 Node 23.6+、npm 和 GPU;相关 packages 会在首次使用时安装(rendering)。提醒用户注意一次性安装,以及当 C: 盘空间不足时需要设置
    $BB_RENDER_HOME
    。
  • Blender、FFmpeg、GPU:按照 查找工具定位并探测。
  • Havok 物理:需要安装 MCP plugin 的 Blockbench desktop,以及 Havok Physics Animations plugin(执行
    blockbench_launch
    后可使用
    havok_*
    tools)。如果不可用,询问是否改用关键帧动画。
  • AI 贴图(PBR):生成 albedo 需要
    FAL_KEY
    。如果缺失,应说明并改用程序化绘制。
  • glTF 路径(glTF route):如果 Blender 没有
    .bbmodel
    importer,则需要 Blockbench desktop 来导出 glTF。

3. Plan

3. 规划

Write
plan.md
in the project folder: the brief, each answer translated with option specs (geometry budget, texel density, texture size, filtering, channels, scene contents, camera and lights, frame range), the part list with real-world sizes, the animation beats per second of the render, and the shot list. Keep it current; subagents read it.
在项目文件夹中编写
plan.md
:包含 brief;按照 option specs 将每个回答转换为具体规格(几何体预算、texel density、贴图尺寸、filtering、通道、场景内容、相机和灯光、帧范围);带真实尺寸的部件清单;每个渲染秒的动画节拍;以及 shot list。持续更新该文件,供 subagents 读取。

4. Build the models

4. 构建模型

Follow headless build: layout, build order, verification after each pass, and parallel subagents for complete scenes and rooms. Textures and animation follow the rows in option specs for the chosen answers. Share each write tool's
web_app
link with the user as models land.
Gate before Blender: every asset validates, its contact sheet and a posed render look right, and the plan's budgets hold. Havok scenes are baked and saved first.
按照 无头构建执行:包括布局、构建顺序、每轮后的验证,以及为完整场景和房间启用并行 subagents。贴图和动画依据所选答案在 option specs 中对应的行。每完成一个模型,就将相应写入 tool 的
web_app
链接分享给用户。
进入 Blender 前的关卡:每个 asset 均通过验证,其 contact sheet 和摆姿渲染看起来正确,并且符合 plan 中的预算。Havok 场景须先完成烘焙并保存。

5. Build the Blender scene

5. 构建 Blender 场景

Follow Blender handoff: import, run the fix pass, assemble, light and stage cameras for the scene type, apply the render preset, and save
blender/<slug>.blend
from scripts kept in
blender/scripts/
.
按照 Blender handoff执行:导入、运行修复流程、组装场景、根据场景类型布光并设置相机、应用渲染预设,并使用保存在
blender/scripts/
中的脚本将场景保存为
blender/<slug>.blend
。

6. Render deliverable

6. 渲染交付物

Copy render_worker.py and render.cmd into
render/
, fill the placeholders, and confirm none remain. Render the verification stills, look at them, fix what they show, and time one full-quality frame. Do not start the full render unless the user asks.
将 render_worker.py 和 render.cmd 复制到
render/
,填入占位符并确认没有遗漏任何占位符。渲染验证静帧,检查这些画面,修正其中显示的问题,并测量一帧完整质量渲染所需的时间。除非用户要求,否则不要启动完整渲染。

7. Report

7. 报告

Reply with:
  • the project folder, the
    .blend
    ,
    render/render.cmd
    and the video path it will write;
  • the verification stills (send or link them);
  • the
    web_app
    links of the final models;
  • GPUs found, seconds per frame, frame count and estimated total render time;
  • what was substituted or skipped (no
    FAL_KEY
    , no Havok, glTF route) and anything unverified.
回复应包含:
  • 项目文件夹、
    .blend
    、
    render/render.cmd
    以及将要写入的视频路径;
  • 验证静帧(发送文件或提供链接);
  • 最终模型的
    web_app
    链接;
  • 检测到的 GPU、每帧耗时、帧数及预计总渲染时间;
  • 进行了哪些替代或跳过了哪些步骤(例如缺少
    FAL_KEY
    、缺少 Havok、采用 glTF 路径),以及哪些内容尚未验证。