blockbench-new-model

Q: How do I install blockbench-new-model?

Run npx skills add https://github.com/jasonjgardner/blockbench-mcp-project --skill blockbench-new-model in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/blockbench-new-model/ (personal) or .claude/skills/blockbench-new-model/ (project).

by jasonjgardnerMedia & Creative

Start a new Blockbench model or scene from a short intake form and deliver a render-ready Blender scene. Asks for quality, texture style (PBR 2K/HD/standard, vanilla Minecraft, classic Blockbench), animation (keyframed or Havok physics), scene type (scene, vignette, isometric room), subject, output folder and render preset; builds .bbmodel files with the headless Blockbench MCP server and parallel agents; imports them into Blender; and writes a render.cmd that renders stills on every GPU and assembles them into a video with FFmpeg. Use when the user runs /blockbench-new-model or asks for a new Blockbench model delivered as a rendered Blender scene or video.

Installs: 168
Last 24h: —
Trending rank: —
Security: Not audited

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/jasonjgardner/blockbench-mcp-project --skill blockbench-new-model

01
Install
Run the command in your project. The CLI asks which agents to install for.
02
Describe the task
Ask your agent for something this skill covers. It loads blockbench-new-model automatically when relevant.
03
Review the result
The agent follows the SKILL.md steps, scripts and references bundled with the skill.

SKILL.md

The instructions your agent reads when the skill is triggered.

Blockbench New Model

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.

1. Intake form

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.

Header	Question	Options (label: description)
Quality	What quality level?	High-quality: detailed silhouette, smooth rounded parts, detail in textures · Low-poly: faceted, bold shapes, flat color regions
Textures	Which 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
Animation	How should it move?	Animated: keyframed clips · Havok physics: physics-based motion baked with the Havok plugin in Blockbench desktop
Scene	What 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.

Header	When	Question	Options
Subject	No argument was given	What is the subject?	Two or three example subjects suited to the round 1 answers; the user types their own through Other
Resolution	Textures = PBR	Which 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
Output	Always	Where should the project folder go?	A parent folder the user chose before, if you know one · Current directory: `./<slug>/` · anything else through Other
Render	Always	Which 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.

2. Preflight

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.

3. Plan

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.

4. Build the models

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.

5. Build the Blender scene

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/.

6. Render deliverable

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.

7. Report

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.

FAQ

What does the blockbench-new-model skill do?

How do I install blockbench-new-model?

Run npx skills add https://github.com/jasonjgardner/blockbench-mcp-project --skill blockbench-new-model in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/blockbench-new-model/ (personal) or .claude/skills/blockbench-new-model/ (project).

Which AI agents can use blockbench-new-model?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is blockbench-new-model safe to install?

No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.

Who maintains blockbench-new-model?

It is published by jasonjgardner in jasonjgardner/blockbench-mcp-project. It has 168 recorded installs.