didibyte-3d
DIDIBYTE plugins for 3D character work
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/itsdidibyte--mascot-engineering)HTML
<a href="/listings/itsdidibyte--mascot-engineering"><img src="/api/listings/itsdidibyte--mascot-engineering/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Python packages (once):
pip install numpy pillow scipy jsonschema scikit-image # scikit-image optional
**As a Claude Code plugin (recommended):** in Claude Code run
/plugin install mascot-engineering --marketplace itsdidibyte/mascot-engineering
(older Claude Code: `/plugin marketplace add itsdidibyte/mascot-engineering`, then `/plugin install mascot-engineering@didibyte-3d`). The skill is then `/mascot-engineering:mascot-engineering` and the agents are `mascot-engineering:<agent>`.
**Without the plugin system:** copy `skills/mascot-engineering` to `~/.claude/skills/` and `agents/*.md` to `~/.claude/agents/` (all projects), or to `.claude/skills/` and `.claude/agents/` inside one project.
Blender 4.2 or newer is needed only for the `.blend` build step (`blender` on PATH, or tell Claude the path).
Check the install (from a clone of this repo):
python3 -m unittest discover -s tests -v # 36 tests, Blender ones skip if not installed BLENDER=/path/to/blender python3 -m unittest tests.test_blender_smoke -v # once per Blender version
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 15 lines
- {
- "name": "didibyte-3d",
- "description": "DIDIBYTE plugins for 3D character work",
- "owner": {
- "name": "DIDIBYTE",
- "url": "https://github.com/itsdidibyte"
- },
- "plugins": [
- {
- "name": "mascot-engineering",
- "source": "./",
- "description": "Measured 2D-to-3D mascot pipeline for Blender (alpha)"
- }
- ]
- }
README.md · README · 136 lines
- # mascot-engineering: measured 2D-to-3D for Claude Code + Blender
- > **Status: alpha (0.2.0-alpha).** The measurement pipeline is tested on synthetic mascots
- > with known 3D truth (36 tests, CI). The Blender build step has **not yet been verified in
- > real Blender**, and the pipeline has not yet been run on real AI-generated turnarounds.
- > Please report what breaks. Community project; not affiliated with Anthropic or the Blender Foundation.
- A Claude Code skill and five subagents that turn your mascot's front, side and back
- pictures into a **measured** 3D blockout in Blender, and **prove it matches** by
- rendering the model from the same angles as your pictures and reporting exactly where
- it is too wide, too narrow or missing a part.
- It started as a ChatGPT-drafted kit, reviewed and rebuilt with Claude. See
- `REVIEW_OF_CHATGPT_V1.md` for what was kept, what was wrong and what changed.
- ## Why this exists
- Prompting "make this mascot in Blender" makes the AI guess proportions from looking at a
- picture, and you go round in circles. This kit:
- 1. measures the silhouettes in your images (pixels, not guesses),
- 2. builds a base body whose front and side outlines match the drawings by construction,
- 3. fits limbs, ears and tails, and finds eyes and other drawn features, from the same pixels,
- 4. renders the model with cameras matched to your images and reports, for example,
- `z 0.39-0.51: model TOO NARROW by 0.18 on BOTH edges` (arms missing),
- 5. tests it against a 3/4 view it was NOT built from (and tunes roundness from it),
- 6. iterates one part at a time, with every change logged, and gives you a .blend with
- camera-matched reference image planes for refining by hand (plus optional .glb).
- ## Install
- Python packages (once):
- ```bash
- pip install numpy pillow scipy jsonschema scikit-image # scikit-image optional
- ```
- **As a Claude Code plugin (recommended):** in Claude Code run
- ```
- /plugin install mascot-engineering --marketplace itsdidibyte/mascot-engineering
- ```
- (older Claude Code: `/plugin marketplace add itsdidibyte/mascot-engineering`, then
- `/plugin install mascot-engineering@didibyte-3d`). The skill is then
- `/mascot-engineering:mascot-engineering` and the agents are `mascot-engineering:<agent>`.
- **Without the plugin system:** copy `skills/mascot-engineering` to `~/.claude/skills/` and
- `agents/*.md` to `~/.claude/agents/` (all projects), or to `.claude/skills/` and
- `.claude/agents/` inside one project.
- Blender 4.2 or newer is needed only for the `.blend` build step (`blender` on PATH, or tell
- Claude the path).
- Check the install (from a clone of this repo):
- ```bash
- python3 -m unittest discover -s tests -v # 36 tests, Blender ones skip if not installed
- BLENDER=/path/to/blender python3 -m unittest tests.test_blender_smoke -v # once per Blender version
- ```
- ## Use
- In Claude Code, from your project:
- ```
- /mascot-engineering:mascot-engineering Make a 3D blockout of my mascot. Front: refs/front.png, side (looking right): refs/side.png, back: refs/back.png. Workspace: work/mymascot
- ```
- (Use `/mascot-engineering` if you copied the skill instead of installing the plugin.)
- Claude will ask you only things it cannot measure:
- - **G1**: if your views disagree (for example, the neck is lower in the side view), which
- view is right for that part. You get `02_views/TURNAROUND.png` to look at.
- - **G2**: approve the part plan (which parts, which primitive, what is mirrored).
- - **G3**: approve the Blender renders, or accept the listed remaining differences.
- ## Try it on the example first
- ```bash
- S=skills/mascot-engineering/scripts
- python3 $S/intake.py work/blobby --id blobby --origin render_3d \
- --view front=examples/blobby/refs/front.png --view right=examples/blobby/refs/side.png \
- --view back=examples/blobby/refs/back.png
- python3 $S/silhouettes.py work/blobby
- python3 $S/consistency.py work/blobby # CONSISTENT
- python3 $S/propose_spec.py work/blobby --id body_hull --despike
- python3 $S/render_preview.py work/blobby
- python3 $S/compare.py work/blobby # reports missing arms, ears, feet
- python3 $S/fit_limb.py work/blobby --id arm_l --x-seed 0.3 --z-range 0.3 0.56 --mirror-id arm_r
- python3 $S/fit_limb.py work/blobby --id ear_l --x-seed 0.15 --z-range 0.85 1.0 --mirror-id ear_r --embed 2
- python3 $S/features.py work/blobby --add eye:1,2 # eyes found by colour, placed on the surface
- python3 $S/render_preview.py work/blobby && python3 $S/compare.py work/blobby # front IoU ~0.98, only feet left
- python3 $S/visual_hull.py work/blobby # instant 3D envelope: 03_geometry/visual_hull.obj
- # held-out 3/4 check:
- python3 $S/intake.py work/blobby --id blobby --add --view three_quarter_left=examples/blobby/refs/three_quarter_left.png
- python3 $S/silhouettes.py work/blobby --view three_quarter_left
- python3 $S/check_three_quarter.py work/blobby
- ```
- `compare.py` exits with code 4 while findings remain; that is expected during iteration.
- Swap in `examples/blobby_inconsistent_side/side_head_too_low.png` as the side view to
- see the consistency check catch an AI-style drawing error (head 0.08 lower in the side view).
- ## What you get per character
- ```
- 00_intake/ 01_references/ 02_views/ 03_geometry/ 04_build/ 05_validation/ 06_surface_motion/
- DECISIONS.md OPEN_ISSUES.md HANDOFF.md manifest.json (+ a ZIP from pack.py)
- ```
- `04_build/build_spec.json` is the single source of truth; the .blend is rebuilt from it.
- ## Honest limits
- - Silhouettes give outlines. The exact cross-section between the front and side views,
- and dents no view shows (eye sockets, mouth), are labelled `proposed` and refined by eye
- or sculpting after the blockout.
- - Inconsistent AI views cannot be made consistent by maths: the kit detects the conflict
- and asks you which view wins.
- - The Blender step was written against the Blender 4.2+ Python API but **has not been run
- in the environment where this kit was built** (no Blender there). Run
- `tests/test_blender_smoke.py` once: it builds the example in Blender and checks that
- Blender's renders match the numpy preview pixel for pixel.
- - A blockout is not a rig, UV layout or final topology. Those are later stages.
- ## Repository layout
- ```
- .claude-plugin/ plugin.json, marketplace.json, icon
- skills/mascot-engineering/ SKILL.md, reference/ (7 docs), schemas/, scripts/ (+ mascot_lib/)
- agents/ 5 subagents
- tests/ 36 tests incl. synthetic ground truth + red-team cases (CI: .github/workflows)
- examples/ synthetic mascot "blobby" turnaround, 3/4 view, inconsistent side view
- REVIEW_OF_CHATGPT_V1.md CHANGELOG.md TEST_RESULTS.md CONTRIBUTING.md LICENSE (MIT)
- ```
- ## Roadmap to 1.0
- 1. Blender smoke test passing on 4.2 LTS and the current release.
- 2. Three real mascots taken from AI turnaround to an approved blockout, with timings.
- 3. Then: tag 1.0.
- ## License
- MIT, see `LICENSE`. Example images are synthetic renders made for this project.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.