Automated AI Media Creation with MagicHour AI and GPT-4 Prompt Optimization

1. Workflow Overview

This workflow automates the creation of AI-generated media assets (images or videos) by integrating MagicHour AI's media generation API with OpenAI's GPT-4 prompt optimization. It accepts HTTP POST requests containing structured parameters, dynamically generates detailed prompts optimized for AI generation, submits these to MagicHour AI's services, polls for completion, and finally downloads and returns the generated media.

Logical blocks:

1.1 Input Reception and Parameter Preparation: Receives webhook POST requests and extracts relevant parameters for image or video generation.
1.2 Prompt Generation with GPT-4: Uses OpenAI GPT-4 to generate detailed, structured prompts suitable for MagicHour AI’s image or video generation APIs.
1.3 Media Generation Request to MagicHour AI: Sends requests to MagicHour AI’s image or video generation endpoints with the generated prompt and user parameters.
1.4 Status Polling and Completion Check: Waits and queries the MagicHour API to check generation status until completion or error.
1.5 Media Download and Response: Downloads the produced media files and responds to the original webhook caller.
1.6 Error Handling: Captures and responds with error details from MagicHour AI or HTTP request failures.

2. Block-by-Block Analysis

2.1 Input Reception and Parameter Preparation

Overview: Handles incoming webhook POST requests, extracts and normalizes user input parameters for downstream processing, and routes flow based on media type (image or video).
Nodes Involved:
- Webhook
- If (checks media type)
- Get Data from Webhook (sets standardized fields for image generation)
- Edit Fields2 (sets standardized fields for video generation)
Node Details:
- Webhook
  - Type: HTTP Webhook (Trigger)
  - Configuration: Accepts POST requests on path /generate-image, allows all origins, synchronous response mode.
  - Inputs: HTTP POST request from external client.
  - Outputs: JSON payload with user parameters.
  - Edge cases: Invalid or missing JSON body, unsupported HTTP methods.
- If (media type check)
  - Type: Conditional node
  - Configuration: Checks if body.type equals "image".
  - Inputs: Webhook output.
  - Outputs: Routes to image processing if true; else video processing.
  - Edge cases: Missing or malformed body.type field.
- Get Data from Webhook
  - Type: Set node
  - Configuration: Extracts and maps nested user input fields into normalized structure for image generation prompt.
  - Inputs: From If node (image branch).
  - Outputs: Standardized JSON for prompt generation with fields like body.prompt, body.parameters.style.tool, body.parameters.image_count.
  - Key expressions: Accesses nested JSON via expressions like {{$json.body.parameters.style.prompt}}.
  - Edge cases: Missing nested fields; no validation beyond presence.
- Edit Fields2
  - Type: Set node
  - Configuration: Similar to above but normalizes parameters for video generation (e.g., duration, resolution).
  - Inputs: From If node (video branch).
  - Outputs: Standardized JSON for video prompt generation.
  - Key expressions: Uses expressions to map inputs from webhook JSON.
  - Edge cases: Missing or invalid video-specific fields.

2.2 Prompt Generation with GPT-4

Overview: Uses OpenAI GPT-4 to generate detailed, concise, and structured prompts from minimal user input for either image or video generation.
Nodes Involved:
- Generate Image Prompt (OpenAI node)
- Generate video Prompt (OpenAI node)
Node Details:
- Generate Image Prompt
  - Type: OpenAI API node (LangChain wrapper)
  - Configuration: Uses GPT-4.1 with temperature 0.5 and topP 1 for consistent, descriptive prompt generation.
  - Inputs: Normalized image parameters from "Get Data from Webhook".
  - Outputs: Single string prompt optimized for text-to-image models.
  - Key expressions: Input prompt content dynamically built from user JSON fields.
  - Version specifics: Requires OpenAI API v1.8+ nodes.
  - Edge cases: Ambiguous or incomplete user inputs cause prompt generation to request clarification (per system prompt). API errors or rate limits possible.
- Generate video Prompt
  - Type: OpenAI API node (LangChain wrapper)
  - Configuration: GPT-4.1 with temperature 0.6 and topP 1, tailored for video prompt generation.
  - Inputs: Normalized video parameters from "Edit Fields2".
  - Outputs: Single string prompt optimized for text-to-video models.
  - Key expressions: Dynamically builds content from user inputs.
  - Edge cases: Same as image prompt generation.

2.3 Media Generation Request to MagicHour AI

Overview: Sends the generated prompt and parameters to MagicHour AI’s image or video generation API endpoints, initiating media creation.
Nodes Involved:
- ai-image-generator (HTTP Request)
- text-to-video (HTTP Request)
Node Details:
- ai-image-generator
  - Type: HTTP Request
  - Configuration: POST to https://api.magichour.ai/v1/ai-image-generator with JSON body including name, image count, orientation, style prompt, and tool. Authentication via HTTP Bearer token credential for MagicHour AI.
  - Inputs: Prompt output from "Generate Image Prompt", parameters from "Get Data from Webhook".
  - Outputs: JSON response containing generation job ID, credits charged, and estimated frame cost.
  - OnError: Continues with error output to allow error handling downstream.
  - Edge cases: API errors, auth failures, invalid parameter format.
- text-to-video
  - Type: HTTP Request
  - Configuration: POST to https://api.magichour.ai/v1/text-to-video with JSON body including name, duration, orientation, resolution, and style prompt. Same authentication as above.
  - Inputs: Prompt from "Generate video Prompt", parameters from "Edit Fields2".
  - Outputs: Similar job ID and cost info.
  - OnError: Continues with error output.
  - Edge cases: Same as image generator.

2.4 Status Polling and Completion Check

Overview: Periodically polls MagicHour AI’s API to check the status of media generation jobs until they reach a terminal state (complete, error, or cancelled).
Nodes Involved:
- Wait (Delay node)
- Get Image Details (HTTP Request)
- If1 (Check image status)
- Get Video Details (HTTP Request)
- If4 (Check video status)
- If2 (Check for valid job ID and credits charged)
- If3 (Branch based on media type to route polling)
- Wait1 (Delay for video polling)
Node Details:
- Wait / Wait1
  - Type: Wait node
  - Configuration: Delays execution between status polls.
  - Inputs: From previous status check or media request nodes.
  - Edge cases: No explicit timeout configured; potential infinite polling if API never returns terminal status.
- Get Image Details
  - Type: HTTP Request
  - Configuration: GET request to MagicHour API to retrieve image generation status using job ID from ai-image-generator. Authenticated via bearer token.
  - Outputs: JSON with status field (e.g., "queued", "rendering", "complete", "error", "cancelled").
  - Edge cases: API errors, invalid IDs.
- If1
  - Type: Conditional
  - Configuration: Checks if image status is one of complete, error, or cancelled to determine if polling should stop and proceed.
  - Outputs: If true, proceeds to download; else waits and polls again.
- Get Video Details
  - Type: HTTP Request
  - Configuration: Similar GET request for video job status.
  - Inputs: From Wait1 after delay.
  - Outputs: Status JSON.
  - Edge cases: Same as image details.
- If4
  - Type: Conditional
  - Configuration: Checks video status terminal states as above.
  - Outputs: If complete, proceed; else poll again.
- If2
  - Type: Conditional
  - Configuration: Validates that job ID is not empty and credits charged is greater than zero before proceeding.
  - Outputs: Routes to success flow or error handling.
  - Edge cases: Zero or empty job ID implies failed request.
- If3
  - Type: Conditional
  - Configuration: Re-checks media type to route polling either to image or video wait nodes. This handles different polling loops for images vs videos.

2.5 Media Download and Response

Overview: After successful generation, downloads the media file(s) from MagicHour AI URLs and sends the file back in the webhook response.
Nodes Involved:
- Download Image (HTTP Request)
- Download Video (HTTP Request)
- Respond to Webhook (Response node)
Node Details:
- Download Image
  - Type: HTTP Request
  - Configuration: GET request to URL provided in MagicHour image project downloads array (downloads[0].url).
  - Inputs: From If1 terminal branch on complete status.
  - Outputs: Binary media data to send back.
  - Edge cases: Download failures, broken URLs.
- Download Video
  - Type: HTTP Request
  - Configuration: Same as image but for video downloads.
  - Inputs: From If4 terminal branch on complete status.
- Respond to Webhook
  - Type: Response node
  - Configuration: Sends the downloaded media back as HTTP response to the original webhook caller.
  - Inputs: Media file from download nodes or error details.
  - Edge cases: Large file sizes may cause timeout or memory issues.

2.6 Error Handling

Overview: Captures error messages and codes from MagicHour API or HTTP request failures and returns structured error information to the webhook caller.
Nodes Involved:
- Get Error Details (Set node)
- Respond to Webhook (Response node)
- Sticky Note4 (documentation)
Node Details:
- Get Error Details
  - Type: Set node
  - Configuration: Extracts error message, code, HTTP status, and finish reason from error output JSON.
  - Inputs: From HTTP request nodes on error branch.
  - Outputs: Structured error JSON.
  - Edge cases: Missing or partial error info.
- Respond to Webhook
  - Sends error JSON back to the caller.
- Sticky Note4
  - Documentation explaining that this error capture only applies to HTTP request errors, not internal generation errors.

3. Summary Table

Node Name	Node Type	Functional Role	Input Node(s)	Output Node(s)	Sticky Note
Webhook	HTTP Webhook	Entry point, receives user requests	None	If
If	Conditional	Routes flow by media type (image/video)	Webhook	Get Data from Webhook, Edit Fields2
Get Data from Webhook	Set	Normalizes image input parameters	If	Generate Image Prompt
Edit Fields2	Set	Normalizes video input parameters	If	Generate video Prompt
Generate Image Prompt	OpenAI API (LangChain)	Generates detailed image prompt	Get Data from Webhook	ai-image-generator
Generate video Prompt	OpenAI API (LangChain)	Generates detailed video prompt	Edit Fields2	text-to-video
ai-image-generator	HTTP Request	Sends image generation request	Generate Image Prompt	If2, Get Error Details
text-to-video	HTTP Request	Sends video generation request	Generate video Prompt	If2, Get Error Details
If2	Conditional	Validates job ID & credits charged	ai-image-generator, text-to-video	If3, (empty branch)
If3	Conditional	Routes to image or video wait nodes	If2	Wait (image), Wait1 (video)
Wait	Wait	Delay between image status polls	If3	Get Image Details
Wait1	Wait	Delay between video status polls	If3	Get Video Details
Get Image Details	HTTP Request	Retrieves image generation status	Wait	If1
Get Video Details	HTTP Request	Retrieves video generation status	Wait1	If4
If1	Conditional	Checks image generation completion	Get Image Details	Download Image, Wait	Sticky Note: Explains image statuses and polling logic
If4	Conditional	Checks video generation completion	Get Video Details	Download Video, Wait1
Download Image	HTTP Request	Downloads generated image file	If1	Respond to Webhook	Sticky Note: GET request to download created image or video file
Download Video	HTTP Request	Downloads generated video file	If4	Respond to Webhook
Respond to Webhook	Respond Node	Sends final media or error response	Download Image, Download Video, Get Error Details	None
Get Error Details	Set	Extracts error info from HTTP failures	ai-image-generator, text-to-video	Respond to Webhook	Sticky Note: Captures HTTP request errors, not internal generation errors
Sticky Note	Sticky Note	Explains image status codes and polling	None	None
Sticky Note1	Sticky Note	POST request format and credential note	None	None	Explains required JSON request body format for MagicHour AI
Sticky Note2	Sticky Note	Explains GET request to download media	None	None
Sticky Note3	Sticky Note	POST request response format example	None	None
Sticky Note4	Sticky Note	Captures HTTP error handling scope	None	None

4. Reproducing the Workflow from Scratch

Create a Webhook node
- Name: Webhook
- Path: /generate-image
- HTTP Method: POST
- Response Mode: responseNode
- Allowed Origins: *
- Purpose: Entry point to receive JSON requests from clients.
Add an If node to check media type
- Name: If
- Condition: Check if {{$json.body.type}} equals "image"
- Routes: True branch for images, False branch for videos.
For Image branch: Create a Set node
- Name: Get Data from Webhook
- Purpose: Normalize the incoming webhook JSON into a structured format for image prompt generation.
- Assignments: Copy fields like body.prompt = $json.body.parameters.style.prompt, body.parameters.style.tool, body.parameters.image_count, body.parameters.orientation, body.parameters.name, body.type, and body.action.
For Video branch: Create a Set node
- Name: Edit Fields2
- Purpose: Normalize webhook parameters for video prompt generation.
- Assignments: Map fields such as body.prompt from style prompt, body.parameters.name, body.parameters.end_seconds, body.parameters.orientation, and body.parameters.resolution.
Create OpenAI node for image prompt generation
- Name: Generate Image Prompt
- Type: OpenAI (LangChain wrapper)
- Model: GPT-4.1
- Temperature: 0.5, TopP: 1
- Input: Use structured JSON from Get Data from Webhook to craft a system prompt and user message that instruct GPT to generate a detailed image prompt.
- Credentials: OpenAI API credentials configured.
Create OpenAI node for video prompt generation
- Name: Generate video Prompt
- Type: OpenAI (LangChain wrapper)
- Model: GPT-4.1
- Temperature: 0.6, TopP: 1
- Input: Use structured JSON from Edit Fields2 for prompt generation similarly.
- Credentials: Same OpenAI API credentials.
Create HTTP Request node for MagicHour AI image generation
- Name: ai-image-generator
- Method: POST
- URL: https://api.magichour.ai/v1/ai-image-generator
- Authentication: HTTP Bearer Auth with MagicHour AI token
- JSON Body: Include name, image_count, orientation, style with prompt from Generate Image Prompt output, and tool from input.
- OnError: Continue with error output to capture errors downstream.
Create HTTP Request node for MagicHour AI video generation
- Name: text-to-video
- Method: POST
- URL: https://api.magichour.ai/v1/text-to-video
- Authentication: HTTP Bearer Auth with MagicHour AI token
- JSON Body: Include name, end_seconds, orientation, resolution, and style prompt from Generate video Prompt.
- OnError: Continue with error output.
Create If node to validate job ID and credits charged
- Name: If2
- Conditions: id field not empty AND credits_charged > 0, else error branch.
Create If node to branch polling by media type
- Name: If3
- Condition: Check if original input type is image or video.
Create Wait node for image polling
- Name: Wait
- Default wait duration (e.g., 5-10 seconds).
Create Wait node for video polling
- Name: Wait1
- Default wait duration (e.g., 5-10 seconds).
Create HTTP Request for getting image status
- Name: Get Image Details
- Method: GET
- URL: https://api.magichour.ai/v1/image-projects/{{job_id}} (replace {{job_id}} dynamically from ai-image-generator output)
- Authentication: Same bearer token.
Create HTTP Request for getting video status
- Name: Get Video Details
- Method: GET
- URL: https://api.magichour.ai/v1/video-projects/{{job_id}}
- Authentication: Same bearer token.
Create If node to check image status terminal states
- Name: If1
- Conditions: status equals complete, error, or cancelled.
Create If node to check video status terminal states
- Name: If4
- Conditions: status equals complete, error, or cancelled.
Create HTTP Request node to download image
- Name: Download Image
- Method: GET
- URL: Use first download URL from Get Image Details response (downloads[0].url).
Create HTTP Request node to download video
- Name: Download Video
- Method: GET
- URL: Use first download URL from Get Video Details response.
Create Set node to extract error details
- Name: Get Error Details
- Extract fields: error.message, error.code, error.status, finish_reason from error output.
Create Respond to Webhook node
- Name: Respond to Webhook
- Sends the final binary media file or error JSON response back to the original caller.
Connect nodes as per logical flow:
- Webhook → If (image/video) →
  - Image: Get Data from Webhook → Generate Image Prompt → ai-image-generator → If2 → If3 → Wait → Get Image Details → If1 → Download Image → Respond to Webhook
  - Video: Edit Fields2 → Generate video Prompt → text-to-video → If2 → If3 → Wait1 → Get Video Details → If4 → Download Video → Respond to Webhook
- Error branches from ai-image-generator and text-to-video → Get Error Details → Respond to Webhook
Add sticky notes to document request formats and status codes for maintainability.

5. General Notes & Resources

Note Content	Context or Link
POST request body must follow JSON structure with fields: `name` (string), `image_count` (number), `orientation` (string), and `style` object including `prompt` (string) and `tool` (string).	Sticky Note1
MagicHour AI image status can be: `queued`, `rendering`, `complete`, `error`, or `cancelled`. Polling loop waits until terminal status before proceeding to download.	Sticky Note (near If1)
GET requests are used to download generated media files after completion.	Sticky Note2
POST response from MagicHour AI includes job `id`, `estimated_frame_cost`, and `credits_charged`.	Sticky Note3
HTTP request error handling captures errors occurring during HTTP calls, but does not capture internal generation errors from MagicHour AI API responses.	Sticky Note4
OpenAI GPT-4 prompt generation enforces strict constraints to avoid hallucinations, emotional tone, or content outside user input.	System prompt in OpenAI nodes
Credentials required: OpenAI API key and MagicHour AI Bearer token. Both must be configured and securely stored in n8n credentials.	Credential references in HTTP Request and OpenAI nodes
This workflow is designed to be extensible to other media types or AI services by following the modular blocks and prompt generation approach.	General recommendation

Disclaimer:
The provided text is exclusively derived from an automated workflow created with n8n, an integration and automation tool. This processing strictly adheres to current content policies and contains no illegal, offensive, or protected elements. All handled data is legal and public.

27 KiB Raw Blame History Unescape Escape