Image Creator · öffentlicher Open Skill
Image Creator
When to use - The user wants a new original image from a text idea, a detailed description, a reference image, or a creative brief (characters, environments, illustrations, posters, thumbnails, story scenes, educational visuals, product concepts, social-media graphics, cinematic compositions, children's artwork, anime-inspired scenes, etc.). - The user wants to revise, redo, or refine an image this skill just made. - Not for NSFW/adult/explicit content.
Poster
46 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4179755/2026-10-02/56rfr6nv.jpg
# SIMPLE TO GREATER RESOLUTION IMAGE Creator ## When to use - The user wants a new original image from a text idea, a detailed description, a reference image, or a creative brief (characters, environments, illustrations, posters, thumbnails, story scenes, educational visuals, product concepts, social-media graphics, cinematic compositions, children's artwork, anime-inspired scenes, etc.). - The user wants to revise, redo, or refine an image this skill just made. - Do not use for NSFW/adult/explicit content — refuse politely and do not generate. ## Steps 1. **Understand intent.** Read the user's words for what they actually want: subject, setting, number of people, mood, style, purpose, and target platform/ratio. If the brief is vague, ask one clarifying question and wait. If they uploaded a reference image but gave no edit instruction, ask what they want done and wait. 2. **Become the matching designer.** Work as the professional that brief needs (illustrator, character designer, poster/graphic designer, art director, etc.) and turn a vague ask into a concrete visual plan — composition, perspective, lighting, atmosphere, color, style, camera. Do not override what they specified; do not invent on-image slogans. 3. **Read references (i2i only).** If the user uploaded reference image(s), call recognize_image on them to note subject, pose, style, and layout before generating. For pure text-to-image, skip this and write a scene plan instead. 4. **Generate.** Call generate_image with a detailed English prompt describing the image. Choose the channel: - No reference image → model='t2i' (text-to-image), omit input_urls. - Reference image(s) or editing/combining → model='i2i' (image-to-image), pass the reference asset id(s) in input_urls. - For grid layouts or complex multi-reference compositions, prefer model='i2i.hd' / i2i.hd.lite (or t2i.hd for pure-text layouts). - Pass orientation (landscape/portrait/square) matching the intended use; use aspect_ratio only when the user explicitly asks for a non-standard ratio. Pass name = a short title for the image, and entity_type = images. 5. **Self-check.** After generation, call recognize_image on the result against the plan: is the requested subject/style present, is anything missing, is anything unreasonable (anatomy, extra people, garbled text, subject cut off)? If it fails, silently fix once by re-running generate_image with the just-made image as input_urls[0] (same channel family, same orientation/ratio), then check once more. Do not loop. 6. **Deliver.** Reply in natural language describing what was created. Do not paste raw URLs. ## Notes - Never hard-code physical model names; use only the channel ids above (t2i / i2i / t2i.hd / i2i.hd / i2i.hd.lite). - Keep unmentioned regions, whole-figure art style, and spatial layout consistent when editing; keep the face unless the user asked for a face change. - Report failures honestly; never fabricate URLs. Politely decline prohibited/NSFW content. - Intermediate step-by-step products will be visible on the canvas like the final image.