Menu
Generate one atomic 3D object from an approved object.json or direct source image.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Generate one clean plate image for a world by removing confirmed foreground objects or user-specified content.
Create, inspect, and manage an image-blaster project envelope under worlds/<slug>. Use before other image-blast skills or whenever the user asks about project state.
Generate world ambience, object impact sounds, or custom sound effects through the FAL ElevenLabs SFX endpoint.
Analyze source images, write image.json, present object candidates, and create object.json files after user confirmation.
Discover and run arbitrary FAL operations that do not fit the narrower image-blast skills.
Generate or resume the static 3D environment for a world using World Labs Marble.
| name | image-blast-3d |
| description | Generate one atomic 3D object from an approved object.json or direct source image. |
Use this skill to generate exactly one object model.
Require a world slug and one clear object id/name. If ambiguous, ask for one atomic physical object.
Inspect worlds/<slug>/output/<object>/ with ls -a.
Resolve the object from object.json, image.json, or source analysis JSON.
If no object.json exists and the object is clear, create the minimal
durable object intent before generation:
{
"schema_version": 1,
"world": "<slug>",
"object": {
"id": "<object-slug>",
"name": "<object name>",
"description": "<literal object description>",
"materials": [],
"source_images": [],
"evidence": [],
"generate_as_3d_object": true,
"working_dir": "worlds/<slug>/output/<object-slug>"
},
"updated_at": "..."
}
Preserve available description, materials, source_images, and
evidence. Do not write generated status, jobs, file lists, or request
lifecycle into object.json.
Build an object-specific extraction prompt that isolates one physical object:
Isolate the <target object> from this image. Reproduce it exactly as shown -- same colors, materials, and proportions. White background, centered, tight crop, studio lighting. No other objects, no scene, no people, no text, no shadows on the ground. Isolate the object and remove all clustered, adjacent, overlapping, or items resting on the target object. Create a clean render of that one single object that is true to the source image.
The prompt must ask for one atomic physical instance only — not a pair, set, cluster, category example, or adjacent duplicate. If the source contains similar objects, identify the target by location and exclude the others.
Generate with:
node .claude/scripts/asset-pipeline/generate-single-asset.mjs --world "<slug>" --object-id "<object-id>" --image-edit-prompt "<object-specific extraction prompt>"
Hunyuan is the default provider with defaults --face-count 50000
(range 40000-1500000), --enable-pbr true, --generate-type Normal.
Use --provider meshy only when the user asks for Meshy. Pass any of
--face-count, --generate-type Normal|LowPoly|Geometry,
--polygon-type triangle|quadrilateral (LowPoly only),
--enable-pbr true|false, or Meshy-specific flags only when the user
asks for that detail/topology.
Use --reference-only for iterative reference refinement (stops after
the image-edit step). Later runs reuse the latest reference PNG by
default, including --regenerate. Use --regenerate-reference only when
the user asks for a fresh extraction from the source image.
Repair missing local files from request metadata when needed:
node .claude/scripts/project/ensure-local-assets.mjs --from "<request-json-path>"
Report object id, output directory, reference image, generated model files, and request metadata.