Long-Form Podcast to Viral Shorts Studio

Transform 60-minute podcast episodes, interview streams, and YouTube webinars into ready-to-publish vertical viral shorts (TikTok, Instagram Reels, YouTube Shorts) with automated multi-speaker tracking, virality scoring, animated karaoke captions, and broadcast-grade audio mastering.

This recipe leverages FotoHUB's Shorts Engine, coordinating Whisper Large-v3 timestamped transcription, Claude 3.5 / DeepSeek virality scoring, computer-vision face tracking with multi-speaker split layouts, semantic B-roll cutaway insertion, and 2-pass EBU R128 loudness normalization (-14 LUFS).

TIP

All API usage in this guide is billed purely in USD. There are no synthetic credits or tokens to manage. You are charged directly against your wallet.available_usd balance.

1. Production Architecture Flowchart

The Shorts Engine orchestrates 10 specialized microservices, utilizing distinct GPU clusters to optimize memory access and inference speed.

mermaid

flowchart TD
    %% Define styles
    classDef ingest fill:#1e40af,stroke:#93c5fd,stroke-width:2px,color:#fff;
    classDef ai fill:#065f46,stroke:#6ee7b7,stroke-width:2px,color:#fff;
    classDef av fill:#9a3412,stroke:#fdba74,stroke-width:2px,color:#fff;
    classDef export fill:#4c1d95,stroke:#c4b5fd,stroke-width:2px,color:#fff;

    subgraph "1. Ingestion Phase"
        A["1. Long-Form Ingest<br/>(YouTube / S3 60-min MP4)"]:::ingest
    end

    subgraph "2. AI Analysis & Segmentation Phase"
        B["2. Whisper Large-v3<br/>(Word-level timestamps & Diarization)"]:::ai
        C["3. Acoustic Prosody & Topic Analysis<br/>(Arousal spikes, speaker turn-taking)"]:::ai
        D["4. Virality Scoring (B-Score)<br/>(Claude 3.5 / DeepSeek)"]:::ai
        E["5. Top Clip Extraction<br/>(15s–60s narrative units)"]:::ai
        
        B --> C
        C --> D
        D --> E
    end

    subgraph "3. Audio/Visual Pipeline"
        F["6. Face-Tracking & Multi-Cam<br/>(Auto camera cut / Stacked Split)"]:::av
        G["7. Word-by-Word Karaoke Captions<br/>(Dynamic ASS glow & emojis)"]:::av
        H["8. Contextual B-Roll Overlay<br/>(Transcript-triggered cutaways)"]:::av
        I["9. 8-Stage Audio Chain<br/>(EBU R128 -14 LUFS)"]:::av
        J["10. NVENC GPU Render<br/>(1080x1920 MP4)"]:::av
        
        F --> G
        G --> H
        H --> I
        I --> J
    end
    
    subgraph "4. Delivery Phase"
        K["Webhook Notification<br/>(HMAC-SHA256 Signed)"]:::export
        L["BYOB Export<br/>(S3 / R2 Bucket Delivery)"]:::export
        
        J --> K
        K --> L
    end

    A --> B
    E --> F

Dedicated GPU workers

Transcription, audio enhancement, optional lip-sync, B-roll matching and the final hardware-encoded render each run on dedicated FOTOhub GPU workers, so stages of one job do not compete for the same hardware.

2. API Endpoints Overview

The complete Shorts Engine lifecycle requires coordinating these REST endpoints:

POST /v1/shorts/clips — Submit a long-form video for asynchronous clipping.
GET /v1/shorts/clips/{job_id} — Poll for status, live progress, and final outputs. There is no separate SSE stream for a single job — poll this endpoint.
POST /v1/ai/transcribe — Standalone endpoint for transcription.

3. Asynchronous Clipping: `POST /v1/shorts/clips`

Submits a 60-minute video for complete end-to-end processing. Because this process can take 3-10 minutes depending on the source length and requested clips, it returns immediately with a 202 Accepted and processes in the background.

Full Parameter Table

Parameter	Type	Required	Default	Description
`source_url`	string	Yes	—	Public MP4 URL, presigned S3 link, or YouTube link to ingest. Max 2048 chars.
`source_type`	string	Yes	—	Type of source: `"url"`, `"youtube"`, `"s3"`, `"gcs"`.
`title`	string	No	`"Untitled"`	Project title for metadata tracking.
`language`	string	No	`"en"`	ISO 639-1 language code (e.g., `"en"`, `"es"`, `"fr"`). Auto-detected if omitted.
`max_clips`	integer	No	`10`	Max clips to discover and render (range: `1`–`30`).
`min_duration`	integer	No	`15`	Minimum duration in seconds (range: `5`–`180`).
`max_duration`	integer	No	`60`	Maximum duration in seconds (range: `10`–`300`).
`aspect_ratio`	string	No	`"9:16"`	Target aspect ratio: `"9:16"`, `"1:1"`, `"4:5"`, or `"16:9"`.
`caption_style`	string	No	`"karaoke"`	Style: `"karaoke"`, `"hormozi"`, `"beasty"`, `"neon"`, `"clean"`, `"none"`.
`captions`	boolean	No	`true`	Whether to burn in captions. Set to `false` if platform captions are preferred.
`reframe`	boolean	No	`true`	Enable CV face-tracking to automatically crop 16:9 to 9:16.
`hooks`	boolean	No	`true`	Auto-generate a 3-second animated text overlay for the hook.
`covers`	boolean	No	`true`	Generate high-CTR title card thumbnails.
`broll`	boolean	No	`true`	Semantically inject B-roll cutaways during long monologue segments.
`enhance_audio`	boolean	No	`true`	Run 8-stage DSP chain (denoise, EQ, limit).
`remove_filler`	boolean	No	`true`	Automatically cut "ums", "ahs", and dead air > 1.2s.
`retention_model`	boolean	No	`true`	Use Claude 3.5 to evaluate pacing specifically for TikTok retention curves.
`webhook_url`	string	No	—	HTTPS endpoint receiving event notifications on completion.
`webhook_secret`	string	No	—	Secret string used to sign HMAC-SHA256 headers.
`reference`	string	No	—	Your internal reference ID (e.g., `user_123_proj_456`).
`settings.multicam`	string	No	`"stack"`	`"stack"` (Top/Bottom split) or `"dynamic"` (active speaker active cut).
`settings.multicam_dim`	float	No	`0.18`	Brightness dimming applied to the inactive speaker (0.0 to 1.0).
`settings.caption_highlight_color`	string	No	`"#FFDF00"`	Hex color for the active karaoke word.
`settings.caption_font`	string	No	`"Montserrat"`	Google Font name for captions.
`settings.broll_style`	string	No	`"cinematic"`	B-roll aesthetic: `"cinematic"`, `"memes"`, `"minimalist"`.
`settings.cover_text`	string	No	`"auto"`	Text to place on the thumbnail, or `"auto"` for AI generation.
`settings.target_platform`	string	No	`"tiktok"`	Tunes safe-zones: `"tiktok"`, `"reels"`, `"shorts"`.

4-Way Orchestration Examples

PythonTypeScriptGocURL

python

import os
import requests
import json

def submit_clipping_job(video_url: str):
    api_key = os.environ.get("FOTOHUB_API_KEY", "fh_live_your_api_key")
    url = "https://apis.fotohub.app/v1/shorts/clips"
    
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json"
    }
    
    payload = {
        "source_url": video_url,
        "source_type": "url",
        "title": "Lex Fridman #333",
        "max_clips": 5,
        "min_duration": 30,
        "max_duration": 60,
        "aspect_ratio": "9:16",
        "caption_style": "hormozi",
        "reframe": True,
        "hooks": True,
        "broll": True,
        "enhance_audio": True,
        "remove_filler": True,
        "webhook_url": "https://api.yourdomain.com/webhooks",
        "webhook_secret": "whsec_super_secret_key",
        "settings": {
            "multicam": "dynamic",
            "caption_highlight_color": "#00FF00",
            "target_platform": "shorts"
        }
    }
    
    response = requests.post(url, headers=headers, json=payload)
    response.raise_for_status()
    
    data = response.json()
    print(f"Accepted Job: {data['job_id']}")
    print(f"Cost Deducted: ${data['charged_usd']} USD")
    return data['job_id']

typescript

import fetch from "node-fetch";

async function submitClippingJob(videoUrl: string): Promise<string> {
  const apiKey = process.env.FOTOHUB_API_KEY || "fh_live_your_api_key";
  const url = "https://apis.fotohub.app/v1/shorts/clips";

  const payload = {
    source_url: videoUrl,
    source_type: "url",
    title: "Lex Fridman #333",
    max_clips: 5,
    min_duration: 30,
    max_duration: 60,
    aspect_ratio: "9:16",
    caption_style: "hormozi",
    reframe: true,
    hooks: true,
    broll: true,
    enhance_audio: true,
    remove_filler: true,
    webhook_url: "https://api.yourdomain.com/webhooks",
    webhook_secret: "whsec_super_secret_key",
    settings: {
      multicam: "dynamic",
      caption_highlight_color: "#00FF00",
      target_platform: "shorts"
    }
  };

  const res = await fetch(url, {
    method: "POST",
    headers: {
      "Authorization": `Bearer ${apiKey}`,
      "Content-Type": "application/json"
    },
    body: JSON.stringify(payload)
  });

  if (!res.ok) {
    throw new Error(`Failed to submit job: ${res.statusText}`);
  }

  const data = await res.json();
  console.log(`Accepted Job: ${data.job_id}`);
  console.log(`Cost Deducted: $${data.charged_usd} USD`);
  
  return data.job_id;
}

package main

import (
	"bytes"
	"encoding/json"
	"fmt"
	"io"
	"net/http"
	"os"
)

func submitClippingJob(videoUrl string) (string, error) {
	apiKey := os.Getenv("FOTOHUB_API_KEY")
	url := "https://apis.fotohub.app/v1/shorts/clips"

	payload := map[string]interface{}{
		"source_url":      videoUrl,
		"source_type":     "url",
		"title":           "Lex Fridman #333",
		"max_clips":       5,
		"min_duration":    30,
		"max_duration":    60,
		"aspect_ratio":    "9:16",
		"caption_style":   "hormozi",
		"reframe":         true,
		"hooks":           true,
		"broll":           true,
		"enhance_audio":   true,
		"remove_filler":   true,
		"webhook_url":     "https://api.yourdomain.com/webhooks",
		"webhook_secret":  "whsec_super_secret_key",
		"settings": map[string]interface{}{
			"multicam":                "dynamic",
			"caption_highlight_color": "#00FF00",
			"target_platform":         "shorts",
		},
	}

	body, _ := json.Marshal(payload)
	req, _ := http.NewRequest("POST", url, bytes.NewBuffer(body))
	req.Header.Set("Authorization", "Bearer "+apiKey)
	req.Header.Set("Content-Type", "application/json")

	client := &http.Client{}
	resp, err := client.Do(req)
	if err != nil {
		return "", err
	}
	defer resp.Body.Close()

	if resp.StatusCode != 202 {
		return "", fmt.Errorf("unexpected status: %s", resp.Status)
	}

	var data map[string]interface{}
	json.NewDecoder(resp.Body).Decode(&data)

	jobID := data["job_id"].(string)
	charged := data["charged_usd"].(float64)
	fmt.Printf("Accepted Job: %s\n", jobID)
	fmt.Printf("Cost Deducted: $%.2f USD\n", charged)

	return jobID, nil
}

bash

curl -X POST https://apis.fotohub.app/v1/shorts/clips \
  -H "Authorization: Bearer fh_live_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "source_url": "https://example.com/podcast.mp4",
    "source_type": "url",
    "title": "Lex Fridman #333",
    "max_clips": 5,
    "min_duration": 30,
    "max_duration": 60,
    "aspect_ratio": "9:16",
    "caption_style": "hormozi",
    "reframe": true,
    "hooks": true,
    "broll": true,
    "enhance_audio": true,
    "remove_filler": true,
    "webhook_url": "https://api.yourdomain.com/webhooks",
    "webhook_secret": "whsec_super_secret_key",
    "settings": {
      "multicam": "dynamic",
      "caption_highlight_color": "#00FF00",
      "target_platform": "shorts"
    }
  }'

4. Polling for Job Status: `GET /v1/shorts/clips/{job_id}`

While webhooks are the recommended approach for production, you can manually poll the API to check job status.

Polling Limits

Do not poll more than once every 5 seconds. Excessive polling will trigger HTTP 429 Rate Limits.

Full Parameter Table

Parameter	Type	Required	Location	Description
`job_id`	string	Yes	Path	The ID returned from the POST request (e.g., `job_948a01bf28`).
`include_metrics`	boolean	No	Query	If true, includes detailed processing timing metrics.

4-Way Code Examples

PythonTypeScriptGocURL

python

import time
import requests

def poll_job_status(job_id: str):
    url = f"https://apis.fotohub.app/v1/shorts/clips/{job_id}"
    headers = {"Authorization": "Bearer fh_live_your_api_key"}
    
    while True:
        resp = requests.get(url, headers=headers)
        data = resp.json()
        
        status = data.get("status")
        pct = data.get("progress_pct", 0)
        
        print(f"Status: {status} ({pct}%)")
        
        if status == "completed":
            print("Finished!")
            return data["clips"]
        elif status in ["failed", "canceled"]:
            raise Exception(f"Job failed: {data.get('error')}")
            
        time.sleep(10)

typescript

async function pollJobStatus(jobId: string) {
  const url = `https://apis.fotohub.app/v1/shorts/clips/${jobId}`;
  
  while (true) {
    const res = await fetch(url, {
      headers: { "Authorization": "Bearer fh_live_your_api_key" }
    });
    const data = await res.json();
    
    console.log(`Status: ${data.status} (${data.progress_pct}%)`);
    
    if (data.status === "completed") {
      return data.clips;
    } else if (data.status === "failed") {
      throw new Error(data.error);
    }
    
    await new Promise(r => setTimeout(r, 10000));
  }
}

package main

import (
	"encoding/json"
	"fmt"
	"net/http"
	"time"
)

func pollJob(jobID string) {
	url := "https://apis.fotohub.app/v1/shorts/clips/" + jobID
	client := &http.Client{}

	for {
		req, _ := http.NewRequest("GET", url, nil)
		req.Header.Set("Authorization", "Bearer fh_live_your_api_key")
		
		resp, _ := client.Do(req)
		var data map[string]interface{}
		json.NewDecoder(resp.Body).Decode(&data)
		
		status := data["status"].(string)
		pct := data["progress_pct"].(float64)
		
		fmt.Printf("Status: %s (%.0f%%)\n", status, pct)
		
		if status == "completed" {
			fmt.Println("Clips ready!")
			break
		}
		time.Sleep(10 * time.Second)
	}
}

bash

# Poll job status
curl -X GET "https://apis.fotohub.app/v1/shorts/clips/job_948a01bf28?include_metrics=true" \
  -H "Authorization: Bearer fh_live_your_api_key"

5. Real-Time UI Monitoring

Corrected 2026-09

This page previously documented a per-job GET /v1/shorts/clips/{job_id}/stream Server-Sent Events endpoint. It does not exist — there is no SSE stream for an individual shorts job anywhere in the API. Build your progress bar by polling GET /v1/shorts/clips/{job_id} (section 4 above), which already returns status and live progress.

python

import time
import httpx

def poll_job_progress(job_id: str):
    url = f"https://apis.fotohub.app/v1/shorts/clips/{job_id}"
    headers = {"Authorization": "Bearer fh_live_your_api_key"}
    with httpx.Client() as client:
        while True:
            data = client.get(url, headers=headers).json()
            print(f"Progress: {data.get('progress')}% - {data.get('status')}")
            if data.get("status") in ("completed", "failed"):
                return data
            time.sleep(3)

6. Delivering clips to your own storage (BYOB)

Corrected 2026-09

This page previously documented a per-job POST /v1/shorts/export call that took a job_id plus raw AWS/R2 credentials in the request body. That endpoint does not exist, and sending long-lived secret keys in a per-call JSON body would be a bad design if it did.

The real BYOB mechanism is account-wide, not per-job: register a destination once with POST /v1/destinations, and everything that API key generates — not just shorts clips — is mirrored there automatically. There is no per-job "export this one clip" call; you either receive the clip URL directly in the poll response, or attach a standing destination up front.

Registering a destination (one-time setup)

bash

curl -X POST https://apis.fotohub.app/v1/destinations \
  -H "Authorization: Bearer fh_live_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Production shorts bucket",
    "providerPreset": "aws",
    "bucketName": "my-production-shorts",
    "region": "us-west-2",
    "accessKeyId": "AKIA...",
    "secretAccessKey": "SECRET...",
    "pathPrefix": "2026/09/"
  }'

For Cloudflare R2, use "providerPreset": "r2" and add "accountId". See Cloud Storage for the full destination reference.

7. Standalone Audio Transcription: `POST /v1/ai/transcribe`

Sometimes you only need the Whisper Large-v3 data (e.g., for custom indexing).

Parameter	Type	Required	Description
`audio_url`	string	Yes	URL to audio/video file.
`diarization`	boolean	No	Tag speakers (e.g., `SPEAKER_00`).
`word_timestamps`	boolean	No	Return ms-level timing.

Code Examples

PythonTypeScriptGocURL

python

import requests

def transcribe_audio(audio_url: str):
    url = "https://apis.fotohub.app/v1/ai/transcribe"
    payload = {
        "audio_url": audio_url,
        "diarization": True,
        "word_timestamps": True
    }
    resp = requests.post(url, headers={"Authorization": "Bearer fh_live_your_api_key"}, json=payload)
    return resp.json()

typescript

// Fetch call for transcription

// Go code for transcription

bash

curl -X POST https://apis.fotohub.app/v1/ai/transcribe \
  -H "Authorization: Bearer fh_live_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{"audio_url": "https://example.com/audio.mp3", "diarization": true, "word_timestamps": true}'

8. The Virality Scoring (B-Score) Formula

Rather than guessing solely from text, FotoHUB combines acoustic prosody measurements with Claude 3.5 / DeepSeek narrative evaluations across 5 weighted dimensions:

$$\text{B-Score} = 0.30 \cdot H + 0.25 \cdot V + 0.20 \cdot E + 0.15 \cdot P + 0.10 \cdot A$$

Dimension	Weight	Detection Method	Description
Hook Power ($H$)	30%	First 3s NLP + Acoustic Pitch	Provocative opening query, controversy, hook words (`secret`, `mistake`, `truth`).
Content Value ($V$)	25%	Claude 3.5 Semantic Analysis	Standalone insight or takeaway that delivers a clear payoff.
Emotional Impact ($E$)	20%	Prosody Energy & Arousal	Voice dynamic range, laughing, vocal conviction, and intensity peaks.
Pacing & Energy ($P$)	15%	Words-per-minute & Pause Ratios	Eliminates dead air and drags, targeting optimal 140–180 WPM tempo.
Visual Appeal ($A$)	10%	Landmark Expression Score	Engaged facial expressions and body language cues.

9. Whisper Large-v3 Transcription Deep-Dive

Audio is demuxed and transcribed with word-level boundary detection. The acoustic model outputs normalized millisecond timestamps (start_ms, end_ms), word confidence scores, and diarization tags separating host (SPEAKER_00) from guest (SPEAKER_01).

json

{
  "word": "Autonomous",
  "start_ms": 12400,
  "end_ms": 13100,
  "confidence": 0.99,
  "speaker": "SPEAKER_01"
}

10. Multi-Speaker Reframing & Visual Cut Logic

When two or more people speak in a landscape frame, the engine provides two reframing strategies:

stack (Split Screen): Stacks Host on top and Guest on bottom in a 9:16 viewport. Inactive speakers are subtly dimmed (dim_inactive: 0.18) so the viewer's gaze automatically tracks the active speaker.
dynamic (Active Speaker Switch): Seamless camera cuts panning to whoever holds the floor with smoothing easing algorithms to prevent jerky jumps on fast interruptions.

11. 8-Stage DSP Audio Chain (-14 LUFS)

To conform to TikTok, YouTube, and Instagram loudness standards, audio passes through an 8-stage DSP pipeline using FFmpeg internal filters:

High-Pass Filter: highpass=f=80 cutoff removes rumble and mic handling noise.
FFT Denoiser: afftdn=nr=12:nf=-25 scrubs room hiss.
Noise Gate: agate=threshold=0.01:ratio=10 removes breathing artifacts.
De-Esser: firequalizer attenuates 4–9 kHz sibilance.
Compressor: acompressor=threshold=-20dB:ratio=4:makeup=2 tightens dynamic range.
Presence EQ: Boosts 2–5 kHz for vocal intelligibility on mobile phone speakers.
Brickwall Limiter: Caps peaks at -0.95 dBFS to eliminate digital clipping.
2-Pass EBU R128 Loudness: Targets integrated -14 LUFS with $\le -1.5\text{ dBTP}$ true peak.

12. Karaoke Caption System (ASS Subtitles)

Captions are generated dynamically using the Advanced SubStation Alpha (.ass) format to provide bounce, glow, and color highlights precisely mapped to the start_ms of each Whisper word. Emoticons are injected semantically.

13. B-Roll Semantic Injection

When the broll: true flag is enabled, the pipeline identifies long contiguous monologues (over 8 seconds). It extracts keyword clusters and queries a massive internal video library to seamlessly inject contextual cutaways that keep the visual scene dynamic.

14. Hook Generation

A 3-second opening hook text overlay is generated using Claude 3.5 to summarize the core controversy or question in the clip, overlaying high-impact yellow/white text in the top 1/3 of the frame to stop the scroll.

15. Cover Thumbnail Generation

High-CTR title card designs are generated for YouTube Shorts and Reels covers. Faces are extracted, background blurred, and high-contrast text is overlaid based on the A/B tested settings.cover_text parameters.

16. Platform-Specific Settings

Platform	Safe Zones	Max Length	Recommended B-Score Focus
TikTok	Avoid right edge (icons)	10 mins	Hook Power (H)
IG Reels	Avoid bottom 20% (caption)	90s	Visual Appeal (A)
YT Shorts	Standard	60s	Content Value (V)

17. Node.js Worker Pool (Batch Processing)

For users processing dozens of episodes, here is a Node.js p-limit implementation to manage concurrency safely:

typescript

import pLimit from "p-limit";

const limit = pLimit(5); // 5 concurrent jobs

const podcastUrls = ["url1", "url2", "url3"]; // up to 100 urls

const jobs = podcastUrls.map(url => 
  limit(() => submitClippingJob(url))
);

await Promise.all(jobs);

18. Python FastAPI Webhook Receiver (HMAC-SHA256)

FotoHUB signs webhook payloads using HMAC-SHA256 in the X-FotoHub-Signature header. Listen for the shorts.clip.rendered event.

python

from fastapi import FastAPI, Request, Header, HTTPException
import hmac
import hashlib

app = FastAPI()
WEBHOOK_SECRET = "whsec_super_secret_key"

@app.post("/webhooks/fotohub")
async def handle_webhook(request: Request, x_fotohub_signature: str = Header(None)):
    payload_bytes = await request.body()
    
    expected_mac = hmac.new(
        WEBHOOK_SECRET.encode("utf-8"), 
        payload_bytes, 
        hashlib.sha256
    ).hexdigest()
    
    if not hmac.compare_digest(expected_mac, x_fotohub_signature):
        raise HTTPException(status_code=400, detail="Invalid signature")
        
    payload = await request.json()
    if payload.get("event") == "shorts.clip.rendered":
        print("Clip ready:", payload["data"]["video_url"])
        
    return {"status": "ok"}

19. Retention Feedback Loop

If you have access to TikTok API analytics, you can feed retention drop-off data back into the pipeline by adjusting the settings.multicam_dim or caption pacing to optimize your next run.

20. Exact USD Pricing Breakdown & ROI

Billing is deducted from your unified prepaid USD wallet balance at job submission time.

Operation	Unit Cost (USD)	Notes
Full Clipping Job (`shorts_clip_job`)	$0.75 / job	Flat rate covering ingest, transcription, virality scoring, multi-speaker reframing, captions, B-roll, audio mastering, and NVENC GPU rendering.
Ingest Only (`shorts_ingest`)	$0.05 / video	Standalone step endpoint
Transcription (`shorts_transcribe`)	$0.05 / video	Standalone WhisperX speech recognition
Scene Detection (`shorts_detect_scenes`)	$0.08 / video	Standalone scene cut detection
Clip Generation (`shorts_generate_clips`)	$0.12 / video	Standalone LLM moment scoring
Smart Reframe (`shorts_reframe`)	$0.08 / clip	Standalone face tracking and 9:16 crop
Karaoke Captions (`shorts_captions`)	$0.05 / clip	Standalone ASS subtitle compilation
GPU Render (`shorts_render`)	$0.12 / clip	Standalone NVENC composite

ROI Calculation Example

Extracting 5 high-converting vertical viral shorts from a 60-minute podcast episode costs $0.75 USD total, or $0.15 per short. If each short achieves 50,000 views on YouTube Shorts at a $0.05 RPM, the return is $2.50 per short, yielding a massive positive ROI on API usage.

21. Error Handling and DLQ Patterns

Use a Dead Letter Queue (DLQ) if the webhook fails to acknowledge a 200 OK. FotoHUB will retry webhooks with exponential backoff for up to 72 hours.

json

{
  "event": "shorts.job.failed",
  "job_id": "job_948a01bf28",
  "error": {
    "code": "transcription_timeout",
    "message": "The audio stream was corrupt."
  }
}

Timeout Risks

Ensure your webhook receiver endpoint responds with 200 OK within 5 seconds. Lengthy processing should be offloaded to a background queue.

22. Exhaustive Troubleshooting & Error Codes

When interacting with the Shorts Engine API, you may encounter the following error codes. This exhaustive list covers every possible failure mode across the 10-stage pipeline.

Ingestion Errors (Phase 1)

Error Code	HTTP	Description	Resolution
`err_ingest_url_unreachable`	400	Source URL returned 404 or timed out.	Ensure URL is public or pre-signed properly.
`err_ingest_format_invalid`	400	File is not MP4/MOV/MKV/WebM.	Convert source video to a supported format.
`err_ingest_duration_short`	400	Video is under 5 minutes.	Shorts engine is optimized for long-form (>5m).

AI Analysis Errors (Phase 2)

Error Code	HTTP	Description	Resolution
`err_whisper_oom`	500	VRAM exceeded during transcription.	Automatic retry on larger GPU instance.
`err_no_speech_detected`	422	Whisper found < 100 words.	Verify audio track isn't silent or purely music.

Audio/Visual Errors (Phase 3)

Error Code	HTTP	Description	Resolution
`err_face_track_lost`	422	No faces detected for > 60s.	Ensure video contains visible speakers for reframing.

(Additional 50 error codes omitted for brevity in this summary, but documented in full API spec)

23. Complete Webhook Payload Examples

Below are the exhaustive, un-truncated JSON payload examples for every event the system emits.

`shorts.job.started`

json

{
  "event": "shorts.job.started",
  "timestamp": "2026-09-06T16:15:02Z",
  "job_id": "job_948a01bf28",
  "data": {
    "source_url": "https://storage.fotohub.app/raw/tech_podcast_ep104.mp4",
    "detected_duration_s": 3612.4
  }
}

`shorts.job.transcribing`

json

{
  "event": "shorts.job.transcribing",
  "timestamp": "2026-09-06T16:16:10Z",
  "job_id": "job_948a01bf28",
  "data": {
    "status": "in_progress",
    "progress_pct": 12,
    "eta_seconds": 245
  }
}

`shorts.clip.rendered` (Extended)

json

{
  "event": "shorts.clip.rendered",
  "timestamp": "2026-09-06T16:18:22Z",
  "job_id": "job_948a01bf28",
  "reference": "podcast_ep104_batch",
  "data": {
    "clip_id": "clip_78a1bc09",
    "virality_score": 93.5,
    "score_breakdown": {
      "hook_power": 96,
      "content_value": 94,
      "emotional_impact": 91,
      "pacing_energy": 90,
      "visual_appeal": 88
    },
    "title": "Why Autonomous AI Agents Will Replace Middle Management",
    "hook": "Stop hiring managers for tasks software can execute in 3 seconds.",
    "duration_s": 44.8,
    "timestamps": { "start_ms": 742000, "end_ms": 786800 },
    "video_url": "https://storage.fotohub.app/shorts/job_948a01bf28/clip_01_9x16.mp4",
    "cover_url": "https://storage.fotohub.app/shorts/job_948a01bf28/cover_01.jpg",
    "metrics": {
      "integrated_lufs": -14.05,
      "true_peak_db": -1.5,
      "filler_words_removed": 6,
      "multicam_layout": "stack"
    },
    "metadata": {
      "faces_detected": 2,
      "broll_injected": 1,
      "broll_timestamps": [
        {"start_ms": 750000, "end_ms": 754000, "tag": "robotics"}
      ]
    }
  }
}

24. Detailed Whisper JSON Output Schema

When hitting POST /v1/ai/transcribe, the engine returns highly granular data.

json

{
  "operation": "transcription",
  "audio_duration_s": 3612.4,
  "language": "en",
  "language_probability": 0.998,
  "segments": [
    {
      "id": 0,
      "seek": 0,
      "start": 0.0,
      "end": 4.5,
      "text": " Welcome back to the AI Summit 2026.",
      "tokens": [50364, 4390, 770, 281, 264, 7318, 12795, 2378, 13, 50589],
      "temperature": 0.0,
      "avg_logprob": -0.154,
      "compression_ratio": 1.1,
      "no_speech_prob": 0.001,
      "speaker": "SPEAKER_00",
      "words": [
        {"word": "Welcome", "start": 0.1, "end": 0.6, "probability": 0.99},
        {"word": "back", "start": 0.6, "end": 0.9, "probability": 0.99},
        {"word": "to", "start": 0.9, "end": 1.1, "probability": 0.99},
        {"word": "the", "start": 1.1, "end": 1.2, "probability": 0.99},
        {"word": "AI", "start": 1.2, "end": 1.8, "probability": 0.99},
        {"word": "Summit", "start": 1.8, "end": 2.5, "probability": 0.99},
        {"word": "2026.", "start": 2.5, "end": 3.8, "probability": 0.98}
      ]
    }
  ]
}

25. Extensive Advanced Audio Customization

For power users, settings.audio exposes raw FFmpeg filter chain parameters.

Parameter	Type	Required	Default	Description
`settings.audio.highpass_freq`	integer	No	`80`	Cutoff frequency in Hz.
`settings.audio.denoise_nr`	integer	No	`12`	Noise reduction factor (1-97).
`settings.audio.compressor_threshold`	float	No	`-20.0`	dB threshold.
`settings.audio.compressor_ratio`	float	No	`4.0`	Compression ratio.
`settings.audio.limiter_ceiling`	float	No	`-0.95`	dBFS true peak cap.

This maps directly to:

bash

ffmpeg -i input.wav -af "highpass=f=80,afftdn=nr=12:nf=-25,agate=threshold=0.01:ratio=10,acompressor=threshold=-20dB:ratio=4:makeup=2,loudnorm=I=-14:TP=-1.5:LRA=11" output.wav

Long-Form Podcast to Viral Shorts Studio ​

1. Production Architecture Flowchart ​

2. API Endpoints Overview ​

3. Asynchronous Clipping: POST /v1/shorts/clips ​

Full Parameter Table ​

4-Way Orchestration Examples ​

4. Polling for Job Status: GET /v1/shorts/clips/{job_id} ​

Full Parameter Table ​

4-Way Code Examples ​

5. Real-Time UI Monitoring ​

6. Delivering clips to your own storage (BYOB) ​

Registering a destination (one-time setup) ​

7. Standalone Audio Transcription: POST /v1/ai/transcribe ​

Code Examples ​

8. The Virality Scoring (B-Score) Formula ​

9. Whisper Large-v3 Transcription Deep-Dive ​

10. Multi-Speaker Reframing & Visual Cut Logic ​

11. 8-Stage DSP Audio Chain (-14 LUFS) ​

12. Karaoke Caption System (ASS Subtitles) ​

13. B-Roll Semantic Injection ​

14. Hook Generation ​

15. Cover Thumbnail Generation ​

16. Platform-Specific Settings ​

17. Node.js Worker Pool (Batch Processing) ​

18. Python FastAPI Webhook Receiver (HMAC-SHA256) ​

19. Retention Feedback Loop ​

20. Exact USD Pricing Breakdown & ROI ​

ROI Calculation Example ​

21. Error Handling and DLQ Patterns ​

22. Exhaustive Troubleshooting & Error Codes ​

Ingestion Errors (Phase 1) ​

AI Analysis Errors (Phase 2) ​

Audio/Visual Errors (Phase 3) ​

23. Complete Webhook Payload Examples ​

shorts.job.started ​

shorts.job.transcribing ​

shorts.clip.rendered (Extended) ​

24. Detailed Whisper JSON Output Schema ​

25. Extensive Advanced Audio Customization ​

Long-Form Podcast to Viral Shorts Studio

1. Production Architecture Flowchart

2. API Endpoints Overview

3. Asynchronous Clipping: `POST /v1/shorts/clips`

Full Parameter Table

4-Way Orchestration Examples

4. Polling for Job Status: `GET /v1/shorts/clips/{job_id}`

Full Parameter Table

4-Way Code Examples

5. Real-Time UI Monitoring

6. Delivering clips to your own storage (BYOB)

Registering a destination (one-time setup)

7. Standalone Audio Transcription: `POST /v1/ai/transcribe`

Code Examples

8. The Virality Scoring (B-Score) Formula

9. Whisper Large-v3 Transcription Deep-Dive

10. Multi-Speaker Reframing & Visual Cut Logic

11. 8-Stage DSP Audio Chain (-14 LUFS)

12. Karaoke Caption System (ASS Subtitles)

13. B-Roll Semantic Injection

14. Hook Generation

15. Cover Thumbnail Generation

16. Platform-Specific Settings

17. Node.js Worker Pool (Batch Processing)

18. Python FastAPI Webhook Receiver (HMAC-SHA256)

19. Retention Feedback Loop

20. Exact USD Pricing Breakdown & ROI

ROI Calculation Example

21. Error Handling and DLQ Patterns

22. Exhaustive Troubleshooting & Error Codes

Ingestion Errors (Phase 1)

AI Analysis Errors (Phase 2)

Audio/Visual Errors (Phase 3)

23. Complete Webhook Payload Examples

`shorts.job.started`

`shorts.job.transcribing`

`shorts.clip.rendered` (Extended)

24. Detailed Whisper JSON Output Schema

25. Extensive Advanced Audio Customization