Krea2 Image Editor · skill pubblica Open Skill

Krea2 Image Editor

Krea2 Image Editor è una Open Skill pubblica su Digen. Aprila per vedere cosa crea questo workflow.

@jimi

Custom

7 2 8

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3325120/2026-10-13/nxq97qgf.jpg

# Krea2 Image Editor

General **image-to-image** editing on the platform default **`i2i`** channel (Krea 2 binding). No project state. The source frame is always a real upload or the last shown edit from this conversation.

**Language:** This skill package is **English only**. Reply to the user in **their** language (match the latest user message). Every `generate_image` `prompt` is **English**.

## When to use

- User uploaded an image and wants a visual change while keeping identity and composition where possible.
- User wants another pass on the **last output** from this skill (stronger change, fix artifacts, different background, etc.).
- User names a concrete edit: relight, recolor, restyle, cleanup, swap background, add/remove an object, change clothes (family-safe), change expression.

**Do not use:** no source image and no prior edit to build on; pure text-to-image from scratch; video; upscaling-only jobs with no edit intent; NSFW, nudity, sexual content, fetish, or violence/gore as the goal.

## Pause and guidance (hard)

Any turn that **stops without a new shown image** must end with:

1. One or two short sentences to the user (their language).
2. **`set_guidance`** in the **same turn**, with **at least two** clickable options. Do not call any tool after `set_guidance`.

This applies to: missing upload; vague edit with no image; **Stop first** refusals; tool `success=false`; safety block. Do not leave the user with text only.

Do **not** call `set_guidance` in the same tool batch as `generate_image`. After the image returns, the next turn is the user-visible reply plus `set_guidance` only (no other tools in that turn).

Buttons must be **clickable objects** with `text` and `action` (not bare strings):

- `send_message`: `content` is the full sentence the tap should send.
- `fill_input`: `content` is a starter prompt the user can edit (use when they need to describe a custom edit).

Offer **2–4** options tied to why you stopped or what they might do next. Translate `text` and `content` into the user’s language. Never put channel names or tool names on buttons.

### Guidance examples (adapt to user language)

| Pause reason | Example buttons |
|---|---|
| Need upload | Upload a photo to edit; Describe the edit you want after you upload |
| Edit too vague | Make the background a sunny beach; Change outfit to a red jacket |
| Refused (NSFW / policy) | Ask for a family-safe edit instead; Upload a different image |
| Generation failed | Try again with the same instructions; Upload a clearer photo |
| Success (shown image) | Stronger edit; Softer change (keep more of original); Change background only; Fix hands or face |

Example JSON shape after success:

```json
{
  "suggested_questions": [
    {
      "text": "Stronger edit",
      "action": {"type": "send_message", "content": "Apply the same edit again but make the change more visible."}
    },
    {
      "text": "Keep more of the original",
      "action": {"type": "send_message", "content": "Redo the edit but keep the original look and identity closer."}
    }
  ]
}
```

## Steps

1. **Source image**
   - Prefer the user’s **latest upload** when present.
   - If there is no new upload but they refer to “this result” / “again” / “more”, use the **last shown image** from this skill in the conversation.
   - No usable source: **pause** — explain that image-to-image needs a photo + **set_guidance** (see table).

2. **Edit brief**
   - Merge the user message into a clear edit goal (what changes, what must stay: face, pose, style, etc.).
   - If the user only uploaded with no text, infer a light enhancement (cleanup, color, subtle polish) or **pause** and ask what to change + **set_guidance**.

3. **`recognize_image`** (source only)
   - `image_url`: source `asset_id` or `uri`.
   - `prompt`: `Briefly: (1) medium — photo vs illustration vs 3D; (2) main subject and framing (landscape/portrait/square); (3) lighting and environment; (4) visible issues (blur, clutter); (5) ide