# hitpop-ai-cover

Adopt this skill: https://proskills.md/skills/other/hitpopai-hitpop-skills-hitpop-ai-cover

Page: https://proskills.md/skills/other/hitpopai-hitpop-skills-hitpop-ai-cover
This document is the public skill bundle for that page. It contains SKILL.md and every helper file the catalog carries. No login is required.

## File: skills/hitpop-ai-cover/SKILL.md
size: 4295
sha: 712fc746cdc68bba85dd9d21ddf4f73689913b48
upstream: https://github.com/hitpopai/hitpop/blob/main/skills/hitpop-ai-cover/SKILL.md

````text
---
name: hitpop-ai-cover
description: "AI song cover generator — replace vocals in any song with a virtual idol's voice using RVC v2 via Replicate API. Supports custom voice models, pitch adjustment, and automatic vocal/instrumental separation. No GPU needed."
version: 0.1.0
metadata:
  openclaw:
    emoji: "🎤"
    requires:
      env:
        - REPLICATE_API_TOKEN
      bins:
        - curl
        - ffmpeg
---

# Hitpop AI Cover — Virtual Idol Voice Replacement

Replace the vocals in any song with your virtual idol's AI voice. Uses RVC v2 via Replicate cloud API — no local GPU needed.

## How It Works

```
Original Song → Vocal/Instrumental Separation → RVC Voice Conversion → Mix Back → AI Cover
```

The Replicate model handles everything in one call: separation + conversion + mixing.

## Step 1: Get a Voice Model

You need an RVC v2 voice model (.zip containing .pth + .index file). Options:

**Option A: Use pre-made models**
Browse 100,000+ models at:
- https://voice-models.com
- https://huggingface.co/QuickWick/Music-AI-Voices

Copy the download URL of the .zip file.

**Option B: Train a custom voice for your virtual idol**
```bash
# Step 1: Create dataset from any YouTube video of the target voice
curl -s -X POST "https://api.replicate.com/v1/predictions" \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "version": "zsxkib/create-rvc-dataset",
    "input": {
      "youtube_url": "YOUTUBE_URL_WITH_TARGET_VOICE"
    }
  }'

# Step 2: Train model with the dataset
curl -s -X POST "https://api.replicate.com/v1/predictions" \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "version": "replicate/train-rvc-model",
    "input": {
      "dataset_url": "DATASET_ZIP_URL_FROM_STEP_1"
    }
  }'
# Output: a .zip file with your custom voice model
```

## Step 2: Generate AI Cover

```bash
curl -s -X POST "https://api.replicate.com/v1/predictions" \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "version": "zsxkib/realistic-voice-cloning",
    "input": {
      "song_input": "SONG_URL_OR_UPLOADED_FILE",
      "rvc_model": "CUSTOM",
      "custom_rvc_model_download_url": "MODEL_ZIP_URL",
      "pitch_change": 0,
      "index_rate": 0.5,
      "filter_radius": 3,
      "rms_mix_rate": 0.25,
      "pitch_detection_algo": "rmvpe",
      "crepe_hop_length": 128,
      "protect": 0.33,
      "main_vol": 0,
      "backup_vol": 0,
      "inst_vol": 0,
      "pitch_change_all": 0,
      "reverb_size": 0.15,
      "reverb_wetness": 0.2,
      "reverb_dryness": 0.8,
      "reverb_damping": 0.7,
      "output_format": "mp3"
    }
  }'

# Poll for result
curl -s "https://api.replicate.com/v1/predictions/$PREDICTION_ID" \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN"
# Wait until status = "succeeded", then download output URL
```

## Key Parameters

| Parameter | Default | Description |
|---|---|---|
| `pitch_change` | 0 | Semitones. +12 = male→female, -12 = female→male, 0 = same |
| `index_rate` | 0.5 | Voice model influence (0.4-0.7 best range) |
| `pitch_detection_algo` | rmvpe | Best for singing. Use `pm` for speech |
| `main_vol` | 0 | AI vocal volume adjustment (dB) |
| `inst_vol` | 0 | Instrumental volume adjustment (dB) |
| `reverb_size` | 0.15 | Reverb room size (0-1) |
| `output_format` | mp3 | mp3 or wav |

## Cost

~$0.034 per song on Replicate (Nvidia T4 GPU, ~3 minutes processing).

## Virtual Idol Voice Setup

For each virtual idol, maintain a voice profile:

```json
{
  "idol_name": "星璃 (Xingli)",
  "voice_model_url": "https://huggingface.co/.../xingli_rvc_v2.zip",
  "pitch_change": 0,
  "index_rate": 0.55,
  "style_notes": "Sweet female, slightly breathy, good at ballads"
}
```

Store in workspace as `idol_voices/{name}.json`.

## Rules

1. **Always use `rmvpe` for singing** — other algorithms produce worse pitch tracking
2. **Keep `index_rate` between 0.4-0.7** — below sounds too generic, above sounds robotic
3. **Adjust pitch for gender** — +12 for male→female, -12 for female→male
4. **Download output immediately** — Replicate URLs expire
5. **Verify output** — listen to the first 10 seconds before using in video
````

## File: README.md
upstream: https://github.com/hitpopai/hitpop/blob/main/README.md

````text
# 🎬 Hitpop

**Video AI Multi-Agent System for OpenClaw, Claude Code, and Cursor.**

9 agents + 24 skills. Say "make me a video" → get a finished video.

[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE)
[![Agents](https://img.shields.io/badge/agents-9-orange.svg)](#agents)
[![Skills](https://img.shields.io/badge/skills-19-blue.svg)](#skills)
[![OpenClaw](https://img.shields.io/badge/OpenClaw-compatible-purple.svg)](https://docs.openclaw.ai)

[Quick Start](#quick-start) •
[Architecture](#architecture) •
[Agents](#agents) •
[Skills](#skills) •
[Workflows](#workflows) •
[Contributing](#contributing)

---

## What Is This?

Hitpop is a multi-agent video production system. Users describe what they want in plain language. A fleet of 9 specialized AI agents collaborates to deliver finished videos — from creative direction to social media publishing.

No video production knowledge required. No code required. Just talk.

```
You:    "Make me a headphone product video for TikTok with voiceover"

Hitpop: Analyzes your request
        → Creative designs visual style
        → Producer generates video from your product image (Vidu Q2)
        → Writer writes narration script
        → Producer adds voiceover (Edge TTS) + subtitles (Whisper) + BGM
        → Critic checks quality (IMPACT score ≥ 20/30)
        → Reviewer final approval
        → Exports TikTok 9:16 format
        → Delivers video + platform copy (title, description, hashtags)
```

Works with **OpenClaw**, **Claude Code**, and **Cursor**.

---

## Quick Start

### 1. Clone

```bash
git clone https://github.com/HitPopAI/hitpop.git
cd hitpop
```

### 2. Set API Key

```bash
export ZHIPU_API_KEY="your-key-from-bigmodel.cn"
```

### 3. Install (OpenClaw)

```bash
chmod +x setup.sh
./setup.sh
```

### 3. Install (Claude Code / Cursor)

```bash
# Claude Code
cp -r skills/* .claude/skills/
cp -r agents/* .claude/skills/

# Cursor
cp -r skills/* .cursor/skills/
```

### 4. Use

Talk to the Main agent:

```
> Make me a product promo video
> Make me a 15-second TikTok about AI tools
> Generate 5 social media videos in batch
> Help me launch this project on GitHub
```

---

## Architecture

```
                         ┌──────────────┐
                         │   👤 User    │
                         └──────┬───────┘
                                │
                    ┌───────────▼───────────┐
                    │  🎬 Main              │
                    │  Chief Strategy Officer│
                    └───────────┬───────────┘
                                │
                         ┌──────▼───────┐
                         │  🧠 Planner  │
                         └──────┬───────┘
                                │
       ┌────────────────────────┼────────────────────────┐
       │                        │                        │
 ┌─────▼─────┐           ┌─────▼─────┐           ┌─────▼─────┐
 │ 💡Creative│◄── ⚔️ ──►│ 🎯 Critic │           │ 📰 Scout  │
 └─────┬─────┘           └─────┬─────┘           └───────────┘
       │                       │
 ┌─────▼─────┐           ┌─────▼─────┐
 │ ✍️ Writer │           │ 💻Producer│──→ 15 Skills
 └─────┬─────┘           └───────────┘
       │
 ┌─────▼─────┐           ┌───────────┐
 │ 🔍Reviewer│           │ 📣Promoter│
 └───────────┘           └───────────┘

Skills (Producer's toolbox):
  Generation:  gen-video · gen-image · comfyui
  Templates:   rendervid · shotstack · json2video · creatomate
  Post-prod:   edit · voiceover · lipsync · subtitle · music · twick
  Distribute:  publish · director · pipeline
```

### Agent Philosophy

Every agent operates as a higher-dimensional intelligence — not a passive tool, but an autonomous thinker with standards and the authority to push back.

| Principle | Meaning |
|---|---|
| **Never execute blindly** | Diagnose before producing. Ask the questions that expose the gap between vague wish and concrete vision. |
| **Never let mediocrity ship** | Would a stranger stop scrolling to watch this? If not, it goes back. |
| **Never confuse activity with progress** | One video that changes minds > ten videos nobody watches. |

### Adversarial Collaboration

| Axis | Agents | Dynamic |
|---|---|---|
| Creativity vs. Taste | 💡 Creative ↔ 🎯 Critic | Bold ideas refined through rigorous evaluation |
| Draft vs. Review | ✍️ Writer ↔ 🔍 Reviewer | Scripts polished through iterative feedback |
| Speed vs. Quality | 💻 Producer ↔ 🎯 Critic | Fast execution held to quality standards |

---

## Agents

| # | Agent | Role | Superpower |
|---|---|---|---|
| 0 | 🎬 **Main** | Chief Strategy Officer | Diagnoses before acting, challenges weak briefs |
| 1 | 🧠 **Planner** | Production Architect | Designs minimum-viable pipelines with failure modes |
| 2 | 💡 **Creative** | Creative Director | Would rather ship nothing than ship boring |
| 3 | 🎯 **Critic** | Quality Sovereign | Absolute veto power, IMPACT scoring (≥20/30 to pass) |
| 4 | 💻 **Producer** | Master Craftsman | Executes all technical work via 24 skills |
| 5 | ✍️ **Writer** | Narrative Architect | "The viewer gives you 1.5 seconds. Earn the next 13.5." |
| 6 | 🔍 **Reviewer** | Audience Proxy | Watches output as a stranger scrolling at 2am |
| 7 | 📰 **Scout** | Intelligence Analyst | Delivers insight, not data. "WHY is it trending?" |
| 8 | 📣 **Promoter** | Growth Strategist | Builds reputation through genuine value (never auto-publishes) |

Each agent has 3 files in `agents/<id>/`:

| File | Purpose |
|---|---|
| `soul.md` | Identity, philosophy, decision principles |
| `agent.md` | Model config, tool permissions, inter-agent protocols |
| `user.md` | Project context, preferences, reference data |

---

## Skills

24 skills that Producer uses as its toolbox. All in `skills/`:

| Category | Skill | Backend | Cost |
|---|---|---|---|
| **Generation** | `hitpop-gen-video` | Vidu Q2 (6 models) via Zhipu API | ¥0.20-0.40 |
| | `hitpop-gen-image` | Seedream 4.0/4.5 via Zhipu API | ¥0.20-0.25 |
| | `hitpop-comfyui` | Local open-source (Flux, Wan2.1) | Free |
| **Templates** | `hitpop-rendervid` | Rendervid (open source) | Free |
| | `hitpop-shotstack` | Shotstack cloud API | Free sandbox |
| | `hitpop-json2video` | JSON2Video API | Free tier |
| | `hitpop-creatomate` | Creatomate API | Free trial |
| **Post-prod** | `hitpop-edit` | FFmpeg | Free |
| | `hitpop-voiceover` | Edge TTS / OpenAI / ElevenLabs | Free+ |
| | `hitpop-lipsync` | Wav2Lip / SadTalker / LivePortrait / HeyGen | Free+ |
| | `hitpop-subtitle` | Whisper | Free |
| | `hitpop-music` | FFmpeg + free libraries | Free |
| | `hitpop-twick` | Twick React SDK | Free |
| **Distribute** | `hitpop-publish` | FFmpeg + platform APIs | Free |
| **Reference** | `hitpop-character-anime + hitpop-character-realistic` | Character 3-view turnaround templates | Free |
| | `hitpop-product-sheet` | Product 5-angle reference + style lock | Free |
| | `hitpop-scene-guide` | Scene prompt templates by content type | Free |
| **Orchestrate** | `hitpop-director` | GLM-5-Turbo routing | ~$0.001 |
| | `hitpop-pipeline` | Checkpoint-based workflow engine | Free |

All Zhipu API calls via `https://open.bigmodel.cn/api/paas/v4/`

---

## Workflows

| Workflow | File | Description |
|---|---|---|
| 📋 Video Pipeline | `workflows/video-pipeline.md` | Full 6-phase production with quality gates |
| 📦 Batch Content | `workflows/batch-content.md` | Template-based bulk video generation |
| 🚀 GitHub Launch | `workflows/github-launch.md` | 7-day community outreach campaign |

### Quality Gates (IMPACT Score)

| Dimension | What Critic Evaluates |
|---|---|
| **I**mpact | Would a stranger stop scrolling? |
| **M**essage | Core message clear in 3 seconds? |
| **P**olish | Any artifacts, glitches, sync issues? |
| **A**udience | Matches target platform and audience? |
| **C**raft | Every frame feels intentional? |
| **T**iming | Pacing feels right? |

Pass: ≥ 20/30. Below 20: redo. Below 15: reject.

---

## Repo Structure

```
hitpop/
├── soul.md         
````
