Image Transfer Visual · सार्वजनिक Open Skill
Image Transfer Visual
Image Transfer Visual Digen पर एक सार्वजनिक Open Skill है। खोलें और जानें कि यह वर्कफ़्लो क्या बनाता है।
Manhua
2 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-19/yg3snbbk.jpg
---
allowed-tools:
- generate_image
- recognize_image
---
# Image Transfer Visual
Turns a short text description into a single finished illustration that belongs to the Universal Image Transfer Tool visual world: one lone figure transporting luggage down a worn, dimly lit industrial passage, rendered in muted cinematic anime style.
There is no structured project state. Keep working notes in the conversation and reuse asset URLs already on the canvas.
## When to use
- The user describes a scene, character, or mood in words and wants a picture of it in this app's look.
- The user wants a text-to-image result ("draw / generate / make me an image of ...") with the transfer-corridor atmosphere.
- The user attaches a photo or illustration and wants it converted into / restyled as this visual world (image-to-image).
- The user asks for a variation, a closer framing, or a different prop while keeping the same style.
## Inputs and outputs
- Input: free text describing the subject, plus optionally one attached reference image.
- Output: one image (single frame, never a collage or grid).
- The user's text supplies **what** is in the frame; this skill supplies **how** it looks, how it is framed, and what must never change.
## Steps
1. **Extract the request.** From the user's message identify: the main subject, the action or prop being carried, any environment emphasis, and the desired mood. If the user gave no subject, ask one short clarifying question instead of guessing.
2. **Read the visual contract.** Call `read_skill_file` for `references/style-and-prompt.md` and `references/step-details.md` before writing any prompt. Do not improvise the look from memory.
3. **Pick the generation path.**
- User supplied an attached or referenced image → call `recognize_image` with an English `prompt` asking for subject, pose, props, framing, palette, lighting, and background structure; then build the prompt and call `generate_image` with `input_urls=[reference image]` and `model='i2i.hd.lite'`.
- Text only → build the prompt and call `generate_image` with `model='t2i'`.
4. **Assemble the prompt** using the template and slot rules in `references/style-and-prompt.md` (subject slot from the user, all style / layout / must-keep / must-avoid blocks fixed). Run the pre-flight checklist in `references/step-details.md` before calling the tool.
5. **Call `generate_image`** with:
- `prompt`: the assembled English prompt;
- `model`: `'t2i'` for text-only, `'i2i.hd.lite'` when a reference image is used;
- `input_urls`: the reference image, only on the image-to-image path;
- `orientation`: `'portrait'` by default (this world is a vertical corridor shot); use `'landscape'` only if the user explicitly asks for a wide framing, `'square'` only if they ask for a square avatar-like crop;
- `resolution`: leave unset for the default; use `'1080P'` only when the user asks for a large, print-like or wallpaper image;
- `name`: a short descriptive title, e.g. `Transfer Corridor - Suitcase Girl`;
- `entity_type`: `'illustrations'`.
6. **Reply in natural language.** One or two sentences describing what was produced and which style element dominates; do not paste long URLs. Offer a next variation only when it is genuinely relevant, and if you do, call `set_guidance` in the same turn as the final text.
## Notes
- If `generate_image` returns `success=false`, say so plainly and do not invent a link or claim a result exists.
- Never add watermarks, signatures, brand logos, or readable signage text beyond generic warning marks; never claim the image contains something it does not.
- Keep the subject singular. If the user asks for a crowd or a group, place them as distant unreadable silhouettes so the lone-figure composition survives.
- Refuse requests for sexual or graphic violent content without calling the generator; video-generation-style adult content rules do not apply here, but the image policy does.
- Do not describe or claim f