What problem does it solve?
This Skill guides responsible image generation and editing by enforcing safety gates, ensuring appropriate use of references, and providing clear prompts for consistent outputs.
Core Features & Use Cases
- Reference-first handling for real people and likenesses, anchoring with reference images and avoiding non-consensual or inappropriate representations.
- Factual grounding and gate: if the request depends on a real-world fact, call web_search first and incorporate verified facts into the prompt; use search_images to obtain visuals, not to fulfill factual queries.
- Prompt ownership and clear edits: own or craft prompts that front-load the subject and describe the required mood, composition, and changes in 2–5 sentences; for edits, describe only the changes.
- Sandbox constraints: no internet access from the bash environment; reuse downloaded visuals via search_images and avoid re-fetching.
- Decision-tree guidance: determine when to render a new image, edit an existing one, or stitch multiple references, using a reference-first workflow for complex cases.
- Reference-first procedure: identify the subject, search for target depictions, select best references, and decide whether to stitch multiple references or edit an image in place.
- Stitching guidance: for multiple references, construct a composite canvas using non-destructive, equal-aspect layouts that preserve all subjects.
Quick Start
Describe the image task and the target style or edits to apply, and the system will generate or edit the image using the reference-first workflow.
