blender-convert-2d-3d
Skill11 starsChanged 36 days agoMITView on GitHub
Reconstruct a Blender relief or full model from supplied imagery with explicit depth and hidden-surface assumptions.
In this repository30 other files it gives its agents.
- AGENTS.mdAGENTS.md
- blender-convert-3d-2dSkill
- blender-create-environmentSkill
- blender-create-greyboxSkill
- blender-create-modelSkill
- blender-create-visual-target-2dSkill
- blender-game-exportSkill
- blender-light-cameraSkill
- blender-materialsSkill
- blender-procedural-geometrySkill
- blender-renderSkill
- blender-review-optimizeSkill
- blender-rig-animateSkill
- blender-setupSkill
- blender-uv-bakeSkill
- blender-convert-2d-3dSkill
- blender-convert-3d-2dSkill
- blender-create-environmentSkill
- blender-create-greyboxSkill
- blender-create-modelSkill
- blender-create-visual-target-2dSkill
- blender-game-exportSkill
- blender-light-cameraSkill
- blender-materialsSkill
- blender-procedural-geometrySkill
- blender-renderSkill
- blender-review-optimizeSkill
- blender-rig-animateSkill
- blender-setupSkill
- blender-uv-bakeSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Verify readiness and inspect the supplied image, resolution, perspective and occlusion.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
blender-convert-2d-3d
Inputs and scope
Accept image path, intended use, relief/full model and scale; require the actual image before matching. Also accept optional setting, action/pose, composition, camera, lighting/mood, palette, user references, output specifications and technical constraints when relevant. State assumptions; ask only when a missing answer materially changes the work. A textual style works without reference images. The shared render-style gallery is optional: use an explicitly supplied path or known checkout, inspect a chosen PNG before drawing conclusions, and never fetch missing images automatically.
Use the separately configured official Blender Lab MCP connection. Discover its tools and confirm read-only scene access before mutations; report a connection blocker instead of switching servers. Preserve unrelated objects and settings, scope new content by named collection/run, and checkpoint before destructive edits. Repeated scripts must replace only owned output or create a separate named run. Do not start paid services without authorization.
Use the existing MCP connection whenever it can perform the operation. On Windows, auxiliary execution MUST remain windowless, including brief flashes, during setup, diagnostics, rendering, export, and verification. Before launching, establish suppression for the complete process chain, including client-owned and dependency-owned startup. Use explicit no-console creation (such as subprocess.CREATE_NO_WINDOW) and captured output for reviewed console-only commands. If any required boundary is unknown or unsupported, do not launch: report BLOCKED with the reason and next action, and continue independent safe work. Never retry through a visible terminal, toggle Blender's system console, or weaken suppression. An absolute executable path, cmd /c, Blender --background, or Start-Process -WindowStyle Hidden alone is not proof of no flashes. Keep exit status, bounded timeouts, useful diagnostics without secrets, and cleanup of owned descendants. Do not hide, minimize, restore, focus, or close the existing Blender editor to suppress auxiliary windows. Preserve fresh editor-readiness checks before screenshots.
Parallel visual target and feedback
At the start of a new top-level 3D task, launch one subagent using blender-create-visual-target-2d with the exact prompt, constraints, supplied references/current-scene preview, unchanged features, task ID and an exclusive target directory. Continue Blender inspection and construction immediately while it prepares the prompt-specific target. Nested skills reuse the same pending/completed worker; do not launch one per skill. The target worker must not mutate Blender or recursively delegate.
When it returns, inspect the target and compare a genuine current preview, prioritize visible gaps, apply in-scope corrections and inspect again against the same target version. Use the requested iteration budget, otherwise up to two focused correction passes; report remaining gaps. Preserve render/export/audit-only scope and existing-asset identity. Follow the target skill's parallel-feedback reference when available. If delegation or image generation is unavailable, use parent-side preparation or a supplied target, disclose any prompt-only fallback, and never claim an unperformed comparison. Pure setup checks and explicit user opt-outs bypass this default.
Workflow
- Verify readiness and inspect the supplied image, resolution, perspective and occlusion.
- Choose relief, layered cutout or full 3D reconstruction based on intended use.
- List observed landmarks and inferred depth/backside geometry.
- Set reference scale and align a matching camera or image plane.
- Choose manual/procedural Blender reconstruction without external paid services by default.
- Block out silhouette and major depth layers.
- Refine visible forms while keeping inferred details editable.
- Apply materials or projections without stretching the reference onto unsupported surfaces.
- Compare the matching view with the source and inspect alternate angles for artifacts.
- Deliver editable model, comparison views and assumptions; do not promise accurate unseen geometry.
Execution notes
One image is underdetermined. Ask for extra views only when essential to the user's accuracy target. A depth relief and a complete watertight asset are different deliverables; do not silently substitute one.
Delivery and evidence
Provide editable source and requested outputs, relevant settings/seed/version, observed checks and remaining limitations. Inspect actual images for visual claims. File existence alone is not proof of quality. Prefer previews before expensive batches; do not overwrite unrelated files. Report unavailable downstream checks as unverified.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

