High-Quality Image Expansion · सार्वजनिक Open Skill
High-Quality Image Expansion
High-Quality Image Expansion Digen पर एक सार्वजनिक Open Skill है। खोलें और जानें कि यह वर्कफ़्लो क्या बनाता है।
Custom
3 1 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/nfaw2zvq.jpg
# High-Quality Image Expansion Take one uploaded reference image and turn it into one high-quality expanded image: the original content stays exactly as it is, and the frame is extended outward with matching color, material and light, so the result reads as a single continuous image — not as a generic newly generated picture. There is no structured project state here: the source image is the uploaded asset, and the output is a generated image asset that appears on the canvas automatically. ## When to use - The user uploads (or already uploaded) an image and asks to generate an image from it: expansion, outpainting, extending the canvas, widening the frame, filling blank space around the subject, continuing the scene beyond the original edges. - The user asks to redo, re-generate or adjust an expansion produced earlier in this conversation (e.g. "make the left side wider", "the seam looks off"). - Do not use this skill when no image is required in the output, or when the user only wants analysis, text, captions or metadata. ## Steps 1. **Confirm input and intent.** Identify the reference image the user uploaded (its asset id / `s3://` uri / https URL) and what should be expanded: which side or sides grow, whether the whole frame grows evenly, target orientation, and whether the original pixels must stay untouched. Ask only if the request is genuinely ambiguous; otherwise assume: **keep the original image intact and in place, extend the scene outward**. 2. **Read the style and prompt conventions before generating.** Call `read_skill_file` with `name='high-quality-image-expansion'` and `path='references/style-and-prompt.md'`, and follow it. Fill the `generate_image` `prompt` strictly with what that file defines — Style, Subject and content, Layout, Must keep, Must avoid — starting from its copyable prompt template. Never call `generate_image` with a generic or improvised prompt without reading that file first. 3. **Call `generate_image`** with: - `prompt`: English, built from `references/style-and-prompt.md` — one single visual theme, the source subject unchanged, and the extended area continuing the same perspective, depth, color, material and lighting so no seam is visible. - `input_urls`: the user's uploaded reference image (asset id / `s3://` uri / https URL). This must be the user's image, never a previously generated asset, unless the user explicitly asks to expand a generated result from this conversation. - `model='i2i.hd.gpt'` — the image-to-image channel for reference-image based generation. Always pass a channel like this; never pass a physical model name. - `orientation`: match the direction of the requested result. Expanding into a wider frame → `landscape`; expanding into a taller frame or keeping a vertical source → `portrait`; expanding evenly around a square source → `square`. If the user gives no direction, use the orientation of the source image. - `resolution`: leave it unset when the user has no requirement (platform default `720P`). Pass `resolution='1080P'` or `'2K'` when the user asks for high quality, print or large-size output. Do not invent pixel sizes, and never hardcode `4K`; out-of-range values are clamped by the platform. - Only pass `aspect_ratio` (e.g. `'3:4'`) if the user explicitly asks for a non-standard ratio; it is an escape hatch, not a default. - `name`: a short canvas title such as `High-Quality Expansion`. - `entity_type='expanded_images'`. 4. **Reply in natural language.** One or two sentences describing what was expanded and which parts were preserved. Do not paste long URLs or raw asset ids, and do not invent links. If the user wants another round, adjust the direction or the style conventions and call `generate_image` again with the original source image. ## Notes - **Never change what must be preserved:** the subject's identity, face, structure, proportions and layout, plus any detail the user or the app description emphasizes. Expansion only adds the