Deco Frame Photo · Open Skill 공개 스킬
Deco Frame Photo
Deco Frame Photo은 Digen의 공개 Open Skill입니다. 열어 보면 이 워크플로가 무엇을 만드는지 알 수 있습니다.
Other
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-21/fk4bfuw4.jpg
--- allowed-tools: - generate_image - recognize_image --- # Deco Frame Photo Turn one uploaded photo into a decorated photo frame picture: the user's own image stays intact inside a heart-shaped pastel frame covered in lace, pearls, satin bows and faceted heart gems, on a soft pink dreamy backdrop. This is an image-to-image job — there is no structured project state; everything lives in the conversation and on the canvas. ## When to use - User uploads a picture and asks for a "deco frame", "heart frame photo", "cute frame", "lace/pearl frame", "photo booth frame" or similar. - User wants a variant of a frame already produced in this conversation (other frame color, other frame shape, with or without a caption). - Do not use for plain retouching, background removal, or generic illustration requests that do not involve a decorative frame. ## Inputs and outputs - **Input:** a reference photo (required) + optional short text (a caption, name, date or short greeting the user wants written on the design). - **Output:** one decorated frame image. The uploaded photo must remain recognizable inside the frame window. ## Steps 1. **Confirm the inputs.** You need the uploaded photo. If no photo was attached, ask for one and stop. Note any caption text the user gave, plus any stated preference (frame color, single frame vs. collage, wording of the text). 2. **Read the photo (optional but recommended for busy images).** Call `recognize_image` with the uploaded image to note the subject's pose, framing, head position, clothing colors and background, so the composite keeps the original look. Skip it when the photo is clear and simple. 3. **Load the visual rules.** Call `read_skill_file` with `references/style-and-prompt.md`. It contains the locked Style / Subject / Layout / Must-keep / Must-avoid rules and the base English prompt. Every prompt you write must obey it. 4. **Load the prompt templates.** Call `read_skill_file` with `references/prompt-templates.md` and pick the template that matches the request (single heart frame by default, collage or caption variants when asked). Fill every placeholder with concrete content from the user's photo and text — never ship a template with placeholders left in. 5. **Load the step checklist.** Call `read_skill_file` with `references/step-details.md` and run its pre-generation checklist (photo usage, text handling, orientation/resolution choice, subject-preservation checks, retry rules) before calling the tool. 6. **Generate.** Call `generate_image` once: - `prompt`: the English prompt you assembled from the two reference files (style block + filled template + negative list). English only. - `input_urls`: the user's uploaded photo (asset id / `s3://` / https as provided). This is the image-to-image source. - `model='i2i.hd.lite'` — the fixed channel for this skill. Never write a physical model name. - `name`: a short canvas title, e.g. `Deco Frame - Pink Heart`. - `entity_type='photo_frames'`. - `orientation='portrait'` by default (fall back to the source photo's orientation only if the user asks for a landscape frame). Pass `resolution='1080P'` when the user wants print/crisp gem detail, otherwise leave it unset. - Keep `batch_size` unset; one frame per call. 7. **Reply.** Describe the result in one or two natural sentences (frame color, shape, caption if any). Do not paste long URLs. Then, in the same turn, call `set_guidance` with concrete next options such as a lavender or white frame version, a three-frame collage, or a caption change. ## Notes - If the user supplies no photo, do not invent a subject: ask for the upload. - Keep the person's identity, facial features, hairstyle, expression, pose and clothing recognizable — the frame is decoration, not a restyle. - Text the user provides must appear verbatim and legibly; never invent captions, names or extra copy that the user did not give. - If `generate_image` returns `success=false`, say so plainly and retry onc