Blender Image To 3d logo

Blender Image To 3d

Community
jangtrinh
blender-image-to-3d

Reconstruct assets in Blender from reference images using only native Blender geometry, local skills, and project workflows; includes fidelity contracts, staged passes, and visual verification.

Overview

Publisherjangtrinh
Repositorydesign-os-3d-blender
Skill nameblender-image-to-3d
Stars
102
Forks
24
Bundled files
1
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by jangtrinh on GitHub. Read the source before you install it.

Installation

Install the Blender Image To 3d AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/jangtrinh/design-os-3d-blender/tree/main/.agents/skills/blender-image-to-3d
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/jangtrinh/design-os-3d-blender.git /tmp/design-os-3d-blender
mkdir -p .claude/skills
cp -r /tmp/design-os-3d-blender/.agents/skills/blender-image-to-3d .claude/skills/blender-image-to-3d
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Blender Image To 3d in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Blender Image To 3d on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Blender Image To 3d is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Image → 3D (v3 — Blender native)

Build from supplied references without new generated/retrieved vendor meshes, hosted model calls, paid credits, or third-party downloads. Load blender-agent-core first and use references/review-rubric.md for visual verdicts.

Phase -1 — Pre-Contract Discovery, Multi-Modal Dossier & Socratic Gate (MANDATORY)

Before writing spec.json or running any Blender construction pass, execute this 4-step pre-contract sequence (knowledge/60-pipeline/visual-fidelity-image-to-3d-matching.md §2):

  1. Phase -1A — Pin Primary Reference Bytes & Internet Datasheet Research (engineering-research.md):
    • Pin all supplied reference images in builds/<slug>/references/ with SHA-256 hashes recorded in source-registry.json.
    • For any mechanical, industrial, or COTS-based product, research the web for manufacturer datasheets and standard part specifications (actuators, gas springs, bearings, fasteners, stroke envelopes) and save builds/<slug>/engineering-research.md.
  2. Phase -1B — Super-Detailed Engineering & CMF Prompt (super-detail-prompt.md):
    • Deconstruct every visible and occluded sub-assembly into builds/<slug>/super-detail-prompt.md: exact millimeter envelopes, asymmetry (e.g., Double-Y vs X-base, rear-offset column), internal ribs/bosses, coaxial pivot coordinates, kinematic invariants, and OpenPBR CMF parameters.
  3. Phase -1C — 4-Sheet Multi-Modal Technical Dossier (secondary_generated_sheet):
    • Generate and pin the 4 technical dossier sheets in builds/<slug>/references/ (subordinate to primary photos/CAD in authority, used to resolve occluded angles and assembly hierarchy before modeling):
      1. 4-View Orthographic Blueprint (Front, Side, Top, Back with millimeter bounding dimensions).
      2. 360° Turnaround Sheet (resolving rear, opposite flank, and underside views).
      3. Mechanical Exploded View (discrete sub-assemblies, pins, bushings, and axial assembly vectors).
      4. Parts Catalog & CMF Matrix Sheet (knolling layout of all individual parts with IDs, dimensions, and material finishes).
  4. Phase -1D — Socratic Gate with Owner (before locking master-contract.md & spec.json):
    • Present the extracted dossier (super-detail-prompt.md, engineering-research.md, and the 4 technical sheets) and ask the owner (ask_question) to confirm:
      1. Target Purpose (AGENTS.md §0b): both (recommended: 3D-printable closed solids + studio media), print, or render-only.
      2. Mechanical & Component Scope: primary structural parts vs full functional mechanical / internal COTS hardware discovery and working kinematics.
      3. Motion & Media Contract: frame rate (30 fps), distinct rendered states (<= 120 frames animatic cap), stroke/exploded choreography, and camera views.
    • Lock builds/<slug>/master-contract.md, design-parameters.json, and spec.json only after the Socratic Gate is answered (or explicitly waived by the owner).

For a mechanical or industrial reference, designate one primary view for visible count, facing, placement, and silhouette; use close-ups for local section and construction. Before choosing functional routing, ports, instrument stations, or connection anatomy, read the applicable project drawings or official installation/technical documents. Record whether each rule is a fact, a design choice, or unknown. CAD can support a profile or architecture only when its match and units are recorded; it never establishes identity, dimensions, selection, or compliance by itself.

For a multi-component site, declare one common scale and shared units/axes. Record sourced physical envelopes separately from display envelopes, protrusions and service allowance; mark proxy/analogue dimensions provisional. Name external ports with their elevation, orientation and role, and verify the modules together before final integration. An illustrative scale does not establish equipment selection or capacity.

Do not promise exact hidden geometry from insufficient evidence. Visible reference fidelity has priority.

Pin local reference bytes with source roles before comparison. An original photo, an owner-supplied dimension, a generated blueprint/360 sheet and a manufacturer drawing have different authority. Preserve conflicts and label local adaptations; later generated views cannot authenticate unseen switch, shaft or PCB geometry. When only inline images are visible, disclose missing byte provenance and obtain the real file rather than inventing a local path or hash.

Phase 0 — Native reconstruction mode

EvidenceNative modeRequired gate
Hard-surface, symmetric, dimensionableParametric/data API + mirrored SubD cagesilhouette and edge-control overlays
Organic continuous formBlender sculpt/curve/SubDmulti-view silhouette and curvature review
Repeated modular partsGeometry Nodes or reusable mesh constructorsinstance count and transform checks
User-selected artifact already stored locallyFrozen scaffold: copy, strip vendor materials, repair/retopo in Blenderprovenance, zero-service-call, topology, and visual audit

If references cannot constrain the requested fidelity, stop at the contract and request the missing views. External generation/retrieval is not a fallback.

Phase 1 — Evidence specification

  • Measure normalized extents and landmark ratios from every reference.
  • List critical visible components; every item maps to a named Blender object or surface region.
  • Define scene graph, symmetry, units, material groups, cameras, and render outputs.
  • Store the plan under plans/; build idempotently under builds/<slug>/.

Phase 2 — Staged construction

  1. Blockout: primary silhouette, proportions, camera match.
  2. Form: shared surfaces, duct/body transitions, bevel hierarchy, panel boundaries.
  3. Detail: functional parts, apertures, fasteners, ribbing, seams.
  4. Materials: local Blender node materials only; neutral diagnostic lighting first.
  5. Polish: production naming, transforms, topology, render settings, outputs.

Use data API/bmesh first. Render and file-save operators are allowed where necessary. Keep components separate when editability or animation requires it.

For a new geometry family, the controller first proves the identity-risk prototype at the target apparent scale. Promote a constructor only after its explicit axes, pivot, ports, accepted parameter range, and regression proof have passed for a second real consumer. Treat an instrument body and its process adapter as separate contracts; a camera-visible head is not placement evidence.

Phase 2.5 — Offline retopology gate (frozen scaffold only)

QRemeshify 1.1.0 may be used only when the user selected an already-local scaffold. First create a manifold bounded input with scripts/prepare-qremeshify-input.py; then run scripts/run-qremeshify.py. The wrapper must pass pinned-library hashes, macOS quarantine approval, a 1k–100k triangle input budget, and a hard wall timeout. For highly fragmented or fused scan/generated scaffolds, start without sharp guidance; dense boundary features can crash the native field stage.

Run tests/blender/qremeshify-candidate-test.py, restore/project materials if needed, then render hero and top comparison sheets with scripts/render-qremeshify-candidate.py. Accept the candidate only when topology improves without dimension, silhouette, aperture, propeller, or feature loss. Otherwise keep the Blender-native repaired scaffold. Prefer retopology before components are fused; whole-object retopology of a monolithic drone can smooth or tear small mechanics even when numeric topology passes. QRemeshify changes topology; it does not repair inaccurate reference geometry.

Phase 3 — Verify every pass

Run numeric checks before image checks. Then generate a camera-matched comparison sheet:

bash
scripts/make-comparison-sheet.sh <reference> <render> <sheet>

Choose one verdict: continue, refine-spec, refine-code, request-input, or stop. Two repeats of the same visible failure require changing the modeling approach, not another parameter tweak.

For every declared critical feature, review the primary and supporting views manually at a comparable apparent scale. Name the feature, expected construction layers, source/object/interface, proof shot, and falsifier. A numeric count, clean topology, or an unreadable distant object is not a visual pass. Check risky bores, coplanar joins, fastener families, lettering, dial marks, and specular edges at the final output density before the full batch. Preserve source, camera/settings, and media hashes; a changed source, camera, or resolution invalidates affected proof.

The HQ coverage checker can validate declared receipt bindings with or without --launch. Its guarded --launch route validates first and then controls only its child render. Direct Blender, MCP, and headless renders bypass that launch guard; record the route accurately and do not imply a universal pre-HQ interception.

For Full HD/social delivery, load knowledge/60-pipeline/native-render-delivery.md. Render the actual saved model. Preflight every shot at final aspect, check curved metal surfaces and legends at delivery density, and inspect original images plus decoded animation. A concept-generation image, thumbnail or enlarged screenshot does not qualify a native Blender render. Changes to normals/materials still invalidate affected shading and media proofs even when vertices are unchanged.

Delivery gate

For repeated reference refinement, use the target-bound critic packet in knowledge/60-pipeline/native-agent-iteration.md. Freeze the target/reference, candidate sources and actual proof captures; ask a separate reviewer for one PASS/FAIL/UNKNOWN observation and actionable correction per contracted feature. scripts/native-review.py checks the declared evidence bindings and missing findings. It does not judge images itself or authenticate reviewer independence. A repeated failed feature calls for a different approach; changing the target starts a new contract revision. The controlled native coupon illustrates a missing detail that passes overall dimensions; it is not a general reconstruction benchmark.

Separate contracted feature acceptance from exact likeness. Retain omitted legends, camera differences, RGB tuning and inferred underside details as visible deltas. Missing critical evidence remains UNKNOWN; never weaken the rubric after seeing a failed candidate. A completed still review does not approve unreviewed motion or the physical/electrical design.

  • Deterministic rebuild succeeds twice with stable counts.
  • Critical components, transforms, topology, materials, camera framing, and render settings pass automated tests.
  • Hero and secondary reference sheets are inspected at output resolution.
  • Opposite, underside, and turntable renders expose no missing surfaces or camera failures.
  • Report remaining visible deltas and inferred surfaces explicitly; do not inflate fidelity.
  • If QRemeshify was used, record version, library hashes, settings, elapsed time, topology metrics, comparison verdict, and fallback result.

No live vendor link may remain in delivery. A user-selected local scaffold must be appended as a copy, repaired into a self-contained asset, provenance-tagged, and verified with external-service-calls = 0.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Blender Image To 3d AI skill do?

Reconstruct assets in Blender from reference images using only native Blender geometry, local skills, and project workflows; includes fidelity contracts, staged passes, and visual verification.

Why use Blender Image To 3d on TypingMind?

Because you install it once and use it with any model. Blender Image To 3d is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Blender Image To 3d in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/jangtrinh/design-os-3d-blender/tree/main/.agents/skills/blender-image-to-3d. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Blender Image To 3d?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Blender Image To 3d?

As many as you like. As long as a model supports skills, you can use Blender Image To 3d with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Blender Image To 3d AI skill free?

It is published on GitHub by jangtrinh. Check the repository for licensing terms. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇