Background Creator
Designbackground-creatorBackground Creator is an Agent Skill for design published in the HorizonRobotics/EmbodiedGen repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Generate background 3D scenes with EmbodiedGen using scene3d-cli. Use this skill whenever users ask to create room/indoor background scenes from text prompts, pre-generate backgrounds for layout-cli, or control scene3d generation quality/runtime with retry, seed, and gs3d settings.
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Inspect the Agent Skill at https://github.com/HorizonRobotics/EmbodiedGen/blob/HEAD/embodied_gen/skills/background-creator/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/background-creator, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
Background Creator
Unified entry for EmbodiedGen background scene generation via scene3d-cli.
When To Use
Use this skill when users want to:
- Generate indoor/background 3D scenes from text prompts.
- Pre-generate scene assets for
layout-cli. - Control
scene3d-cliruntime/quality via seed, retry, andgs3dsettings.
Routing Rule (Core)
Use scene3d-cli when input is scene-level text prompts and output target is a background scene (mesh + 3DGS), not single foreground assets.
Pre-checks
- Run commands from the repository root.
- Confirm the active environment is
embodiedgen. - Install scene3d dependencies first if needed:
bash install.sh scene3d - If CLI commands are unavailable, run
pip install -e .to register entrypoints.
Standard Command Template
scene3d-cli --prompts "Art studio with easel and canvas" \
--output_dir outputs/bg_scenes \
--seed 0 \
--gs3d.max_steps 4000 \
--disable_pano_check
Common Parameters
--prompts: one or more scene text prompts.--output_dir: output root directory for generated scenes.--seed: random seed for reproducibility.--n_retry: panorama generation retries.--real_height: force target real-world room height in meters.--pano_image_only: generate only panorama image (debug/fast validation).--disable_pano_check: skip panorama quality check.--keep_middle_result: keep intermediate training artifacts.--gs3d.max_steps: training steps for 3DGS optimization.
Output Conventions
Each prompt is saved under <output_dir>/scene_xxxx/, typically including:
gs_model.plymesh_model.plypano_image.pngprompt.txtvideo.mp4gsplat_cfg.yml
Runtime Expectations
- Typical full generation time is around 30 minutes per scene.
- Use
--pano_image_onlyfor quick prompt validation before full generation.
Failure Handling and Retry
- OOM or GPU pressure: reduce concurrency and lower
--gs3d.max_steps. - Unstable scene quality: increase
--n_retryor adjust prompt specificity. - Missing outputs: verify
--output_dirpermissions and use absolute paths.