Social Compliance Generator v2.0

This project is part of Code & Context's AI Drop of the Week.

This bot posts four times a day to X.com. See the results here.

Overview

Very often, I feel pressured to create content for social media platforms like X, Facebook, and Instagram. However, I hate the hassle of posting to those platforms, so I just don't ever get around to it.

This project aims to address this challenge by using AI to generate daily content for social media platforms, and it gives me a chance to play with various AI APIs and models.

I'm open sourcing the project so you can adapt it and become socially compliant too!

What's New in v2.0

This version has been completely modernized with 2025 AI tools:

Component	v1.0 (2024)	v2.0 (2025)
Scheduler	Trigger.dev	GitHub Actions (cron)
Content Source	Random topic list	Live AI news search (Tavily)
LLM	Claude Haiku (hardcoded)	OpenRouter (configurable)
Image Generation	OpenAI DALL-E 3	Google Nano Banana Pro
Audio Generation	HuggingFace MusicGen	Built into Sora video
Video Generation	FFmpeg (image + audio)	OpenAI Sora 2
X API	v1.1 + v2 hybrid	OAuth 2.0 only

How it Works

A GitHub Actions workflow runs 4 times daily. Upon invocation, it:

1. Searches for AI news using Tavily API to find the latest AI/ML news
2. Generates content using OpenRouter (configurable LLM) to select the most interesting topic and write a post
3. Creates an image using Google Nano Banana Pro (Gemini 3 Pro Image)
4. Generates a video using OpenAI Sora 2 from the image with synchronized audio
5. Posts to X using the X API v2 with OAuth 2.0

Tavily (AI news) → OpenRouter LLM (content) → Nano Banana Pro (image)
                                                      ↓
                    X API v2 (post) ← Sora 2 (video with audio)

Prerequisites

You'll need accounts and API keys for:

1. OpenRouter - LLM access (supports 200+ models)
2. Tavily - AI-powered web search
3. Google Cloud / Vertex AI - Gemini image generation (see Image Generation Setup below)
4. OpenAI Platform - Sora 2 video generation (requires API access)
5. X Developer Portal - OAuth 2.0 credentials
6. Cloudflare - KV storage for OAuth refresh tokens
7. Cloudflare R2 (Optional) - Object storage for workflow persistence and replay

See README-AUTH.md for detailed setup instructions.

Quick Start

1. Clone and Install

git clone https://github.com/intertwine/social-compliance-generator.git
cd social-compliance-generator
npm install

2. Configure Environment

cp .env.example .env
# Edit .env with your API keys

3. Test Locally

npm run generate

4. Deploy to GitHub Actions

1. Go to your repository's Settings → Secrets and variables → Actions

2. Add the following secrets:

   • OPENROUTER_API_KEY
   • OPENROUTER_MODEL (optional, defaults to anthropic/claude-sonnet-4.5-20250929)
   • TAVILY_API_KEY
   • GOOGLE_CLOUD_PROJECT - Your Google Cloud project ID
   • GOOGLE_CLOUD_CREDENTIALS - Service account JSON key (see Image Generation Setup)
   • OPENAI_API_KEY
   • X_API_CLIENT_ID
   • X_API_CLIENT_SECRET
   • X_API_ACCESS_TOKEN
   • X_API_REFRESH_TOKEN
   • CLOUDFLARE_ACCOUNT_ID - Your Cloudflare account ID
   • CLOUDFLARE_KV_NAMESPACE_ID - KV namespace ID for token storage
   • CLOUDFLARE_KV_API_TOKEN - API token with KV write permissions
   • CLOUDFLARE_R2_BUCKET (optional) - R2 bucket name for workflow storage
   • CLOUDFLARE_R2_ACCESS_KEY_ID (optional) - R2 API access key
   • CLOUDFLARE_R2_SECRET_ACCESS_KEY (optional) - R2 API secret key

3. The workflow will run automatically at 6am, 12pm, 6pm, and midnight UTC

4. You can also trigger it manually from the Actions tab

Project Structure

social-compliance-generator/
├── .github/workflows/
│   └── generate-post.yml      # Cron-triggered GitHub Action
├── src/
│   ├── index.ts               # Main orchestration
│   ├── replay.ts              # Workflow replay utility
│   ├── services/
│   │   ├── search.ts          # Tavily web search
│   │   ├── llm.ts             # OpenRouter LLM
│   │   ├── image.ts           # Google Nano Banana Pro
│   │   ├── video.ts           # OpenAI Sora 2
│   │   ├── x.ts               # X API posting
│   │   ├── token-storage.ts   # Cloudflare KV token storage
│   │   └── workflow-storage.ts # Cloudflare R2 workflow storage
│   └── types/
│       └── workflow.ts        # Workflow data types
├── .env.example               # Environment template
├── package.json
└── tsconfig.json

Customization

Change the LLM Model

Set OPENROUTER_MODEL in your environment to any model from OpenRouter's catalog:

OPENROUTER_MODEL=openai/gpt-4o
OPENROUTER_MODEL=meta-llama/llama-3.1-70b-instruct
OPENROUTER_MODEL=google/gemini-pro-1.5

Change Post Frequency

Edit .github/workflows/generate-post.yml and modify the cron schedule:

schedule:
  - cron: '0 */6 * * *'  # Every 6 hours
  - cron: '0 9 * * *'    # Once daily at 9am UTC

Customize Post Tags

Edit src/services/x.ts to change the hashtags and links:

const POST_TAGS = ["YourTag1", "YourTag2"];
const POST_LINKS = [
  { title: "Your Link", url: "https://your-url.com" }
];

Image Generation Setup

The image service uses Google's Gemini models (Nano Banana Pro / Gemini 2.5 Flash) with automatic fallback on rate limits.

Option 1: Vertex AI (Recommended for Production)

Vertex AI provides higher rate limits and better reliability. Required for GitHub Actions.

1. Create a Google Cloud project at console.cloud.google.com

2. Enable billing for the project:

   gcloud billing projects link YOUR_PROJECT_ID --billing-account=YOUR_BILLING_ACCOUNT

3. Enable the Vertex AI API:

   gcloud services enable aiplatform.googleapis.com --project=YOUR_PROJECT_ID

4. Create a service account:

   gcloud iam service-accounts create github-vertex-ai \
     --project=YOUR_PROJECT_ID \
     --display-name="GitHub Actions Vertex AI"
   
   gcloud projects add-iam-policy-binding YOUR_PROJECT_ID \
     --member="serviceAccount:github-vertex-ai@YOUR_PROJECT_ID.iam.gserviceaccount.com" \
     --role="roles/aiplatform.user"

5. Create and download the key:

   gcloud iam service-accounts keys create key.json \
     --iam-account=github-vertex-ai@YOUR_PROJECT_ID.iam.gserviceaccount.com

6. Add to GitHub Secrets:

   • GOOGLE_CLOUD_PROJECT: Your project ID
   • GOOGLE_CLOUD_CREDENTIALS: Contents of key.json

7. Delete the local key file:

   rm key.json

Option 2: Gemini Developer API (Local Development)

For local development, you can use the simpler Gemini Developer API:

1. Get an API key from Google AI Studio
2. Set GOOGLE_API_KEY in your .env file

Note: The free tier has strict rate limits. For production use, set up Vertex AI.

Workflow Storage Setup (Optional)

The workflow storage feature uses Cloudflare R2 to persist intermediate results (news search, content, images, videos) from each workflow run. This enables:

• Replay failed posts: If X posting fails, you can manually repost later
• Debugging: Inspect the full workflow state for any run
• Audit trail: Keep a history of all generated content

Setting Up R2 Storage

1. Create an R2 bucket at Cloudflare Dashboard → R2 → Create bucket

2. Create an R2 API token:

   • Go to Dashboard → R2 → Manage R2 API Tokens → Create API token
   • Select "Object Read & Write" permission
   • Apply to your specific bucket or all buckets
   • Copy the Access Key ID and Secret Access Key

3. Add to your environment (.env for local, GitHub Secrets for Actions):

   CLOUDFLARE_R2_BUCKET=your-bucket-name
   CLOUDFLARE_R2_ACCESS_KEY_ID=your-access-key-id
   CLOUDFLARE_R2_SECRET_ACCESS_KEY=your-secret-access-key

Note: R2 storage is optional. If not configured, workflows will run normally without persistence.

Replay Utility

The replay utility allows you to manage workflow runs stored in R2:

# List recent workflow runs
npm run replay list

# Show details of a specific run
npm run replay show <runId>

# Repost a failed workflow to X
npm run replay post <runId>

# Delete a workflow run from storage
npm run replay delete <runId>

Example output from npm run replay list:

[OK] run-20241115-103000-abc123
    Started: 2024-11-15T10:30:00.000Z
    Completed: 2024-11-15T10:32:15.000Z
    Topic: OpenAI announces GPT-5
    Post ID: 1234567890 (video)

[FAIL] run-20241115-063000-def456
    Started: 2024-11-15T06:30:00.000Z
    Topic: Google releases Gemini 3

Troubleshooting

Video generation fails

The Sora 2 API requires explicit invitation from OpenAI. If video generation fails, the system will automatically fall back to posting an image-only post.

Token refresh errors

Ensure your Cloudflare KV namespace is set up correctly and the API token has write permissions. The initial OAuth tokens should be set in your environment variables (X_API_ACCESS_TOKEN and X_API_REFRESH_TOKEN).

Image generation rate limits

The image service automatically falls back from Nano Banana Pro to Gemini 2.5 Flash on rate limit errors. If both fail:

• Verify billing is enabled: gcloud billing projects describe YOUR_PROJECT_ID
• Check the error message - free_tier means billing isn't properly linked
• Wait a few minutes if you just enabled billing/APIs

R2 storage errors

If you're seeing R2-related errors:

• Verify all three R2 environment variables are set: CLOUDFLARE_R2_BUCKET, CLOUDFLARE_R2_ACCESS_KEY_ID, CLOUDFLARE_R2_SECRET_ACCESS_KEY
• Ensure the API token has "Object Read & Write" permissions
• Check that the bucket name matches exactly (case-sensitive)
• R2 storage is optional - workflows will continue without it if not configured

X video upload issues

The X API v2 video upload uses chunked uploads with dedicated endpoints (as of January 2025):

• 413 Payload Too Large: Videos are uploaded in 1MB chunks. If you see this error, the chunk size may need adjustment.
• Invalid media IDs: After upload, videos need processing time (10-60 seconds depending on size). The system waits automatically.
• Rate limits: Free tier has low limits (17 initialize/finalize per 24h, 85 appends). Consider upgrading your X API tier for production use.

Technical details:

• Endpoints: /2/media/upload/initialize, /{id}/append, /{id}/finalize
• Media category: amplify_video (required for video uploads)
• Authentication: OAuth 2.0 with media.write scope

Rate limits

OpenRouter and other APIs have rate limits. If you're hitting limits, consider:

• Reducing post frequency
• Using a different LLM model
• Adding retry logic with exponential backoff

License

MIT

---

For more fun AI projects and tools, subscribe to the AI Drop of the Week Newsletter.