ugc-video-auto

Este extracto es una copia guardada durante la revisión. La fuente externa contiene la versión completa y actual.

---
name: ugc-video-auto
description: Full UGC ad video pipeline — generates a character image on Higgsfield, then creates a Seedance 2.0 video from it. Orchestrates the complete flow from image prompt to finished UGC video. Use when the user wants to create a UGC ad video end-to-end, or wants to take a generated image from Higgsfield's image tab into video creation. Triggers on "create a UGC video", "make a UGC ad", "UGC pipeline", "turn this into a UGC video", or any request combining image generation with video creation for ads/UGC content. Requires Playwright MCP tools.
---

# UGC Video Auto — Full Pipeline via Playwright

This skill orchestrates the complete UGC ad video pipeline on Higgsfield: generate a character image → select it for video → add a UGC video prompt → generate a Seedance 2.0 video. All automated via Playwright.

---

## Pipeline Overview

```
Step 1: Generate character image  →  /image/soul-v2
Step 2: Navigate to video page    →  /create/video?model=seedance_2_0
Step 3: Select image from gallery →  Upload dialog → "Image Generations" tab
Step 4: Write UGC video prompt    →  Lexical editor (contenteditable)
Step 5: Configure & generate      →  Click Generate
```

---

## Prerequisites

- User must be **logged in** to higgsfield.ai in the Playwright browser session
- **Playwright MCP** plugin must be enabled
- Load each Playwright tool with `ToolSearch` before first use

## Input Requirements

The user provides:
1. **Character description** OR a pre-made image prompt (invoke `/ugc-hot-girl` to generate one)
2. **Product/brand context** — what the UGC video is advertising
3. **Video style** — testimonial, product showcase, unboxing, reaction, etc.

Optional overrides:
- **Image model**: Soul 2.0 (default, best for portraits)
- **Video model**: Seedance 2.0 (default)
- **Duration**: 5s, 8s (default), 10s
- **Ratio**: Auto, 16:9, 9:16 (TikTok/Reels), 1:1
- **Resolution**: 720p (default), 1080p

---

## Full Automation Flow

### Phase 1: Generate the Character Image

#### Step 1.1: Navigate to image generation

```
browser_navigate → url: "https://higgsfield.ai/image/soul-v2"
```

Wait for page load, take a snapshot.

#### Step 1.2: Enter the image prompt

Find the textbox (Playwright ID: `hf:tour-image-prompt`, placeholder: "Describe the scene you imagine"):

```
browser_click → ref: <textbox_ref>, element: "Image prompt textbox"
browser_type → ref: <textbox_ref>, text: "<character image prompt>"
```

If the user hasn't provided an image prompt, use `/ugc-hot-girl` to generate one first.

#### Step 1.3: Generate the image

Find the Generate button (shows "Generate XXXX free gens left"):

**ASK USER FOR CONFIRMATION** — this uses credits/free generations.

```
browser_click → ref: <generate_button_ref>, element: "Generate button"
```

#### Step 1.4: Wait for generation

```
browser_wait_for → time: 20
```

Take a screenshot. Verify the new image appears as the **first item** in the History grid. Confirm by checking the first grid element's innerHTML contains an `<img>` tag with a fresh asset URL.

---

### Phase 2: Create the Video from the Generated Image

#### Step 2.1: Navigate to Seedance 2.0 video page

```
browser_navigate → url: "https://higgsfield.ai/create/video?model=seedance_2_0"
```

Wait for page load, take a snapshot. Confirm "Seedance 2.0" appears in the Model button.

#### Step 2.2: Open the upload media dialog

Find the upload area — look for `generic [cursor=pointer]` containing:
- `paragraph: Upload media`
- `paragraph: Image, Video or Audio`

```
browser_click → ref: <upload_area_ref>, element: "Upload media area"
```

This opens a **dialog** with tabs: Uploads, Image Generations, Video Generations, Elements, Liked.

#### Step 2.3: Switch to Image Generations tab

```
browser_click → ref: <image_gen_tab_ref>, element: "Image Generations tab"
```

Look for `button "Image Generations"` in the dialog's tab bar.

#### Step 2.4: Select the generated image

The most recently generated image is th