Generate AI videos from text or images with Sora 2/Pro & GPT-5 enhancement

Harness OpenAI's Sora 2 for instant video creation from text or images using fal.ai's API—powered by GPT-5 for refined prompts that ensure cinematic quality. This template processes form submissions, intelligently routes to text-to-video (with mandatory prompt enhancement) or image-to-video modes, and polls for completion before redirecting to your generated clip.

📋 What This Template Does

Users submit prompts, aspect ratios (9:16 or 16:9), models (sora-2 or pro), durations (4s, 8s, or 12s), and optional images via a web form. For text-to-video, GPT-5 automatically refines the prompt for optimal Sora 2 results; image mode uses the raw input. It calls one of four fal.ai endpoints (text-to-video, text-to-video/pro, image-to-video, image-to-video/pro), then loops every 60s to check status until the video is ready.

• Handles dual modes: Text (with GPT-5 enhancement) or image-seeded generation
• Supports pro upgrades for higher fidelity and longer clips
• Auto-uploads images to a temp host and polls asynchronously for hands-free results
• Redirects directly to the final video URL on completion

🔧 Prerequisites

• n8n instance with HTTP Request and LangChain nodes enabled
• fal.ai account for Sora 2 API access
• OpenAI account for GPT-5 prompt refinement

🔑 Required Credentials

fal.ai API Setup

1. Sign up at fal.ai and navigate to Dashboard → API Keys
2. Generate a new key with "sora-2" permissions (full access recommended)
3. In n8n, create "Header Auth" credential: Name it "fal.ai", set Header Name to "Authorization", Value to "Key [Your API Key]"

OpenAI API Setup

1. Log in at platform.openai.com → API Keys (top-right profile menu)
2. Click "Create new secret key" and copy it (store securely)
3. In n8n, add "OpenAI API" credential: Paste key, select GPT-5 model in the LLM node

⚙️ Configuration Steps

1. Import the workflow JSON into your n8n instance via Settings → Import from File
2. Assign fal.ai and OpenAI credentials to the relevant HTTP Request and LLM nodes
3. Activate the workflow—the form URL auto-generates in the trigger node
4. Test by submitting a sample prompt (e.g., "A cat chasing a laser"); monitor executions for video output
5. Adjust polling wait (60s node) for longer generations if needed

🎯 Use Cases

• Social Media Teams: Generate 9:16 vertical Reels from text ideas, like quick product animations enhanced by GPT-5 for professional polish
• Content Marketers: Animate uploaded images into 8s promo clips, e.g., turning a static ad graphic into a dynamic story for email campaigns
• Educators and Trainers: Create 4s explainer videos from outlines, such as historical reenactments, using pro mode for detailed visuals
• App Developers: Embed as a backend service to process user prompts into Sora 2 videos on-demand for creative tools

⚠️ Troubleshooting

• API quota exceeded: Check fal.ai dashboard for usage limits; upgrade to pro tier or extend polling waits
• Prompt refinement fails: Ensure GPT-5 credential is set and output matches JSON schema—test LLM node independently
• Image upload errors: Confirm file is JPG/PNG under 10MB; verify tmpfiles.org endpoint with a manual curl test
• Endless polling loop: Add an IF node after 10 checks to timeout; increase wait to 120s for 12s pro generations