Creator Pipeline Glossary

View-velocity metric: views ÷ hours since publish, normalised so 10,000 v/h saturates the 0–1 score. ViralMint's scout uses VPH to rank trending videos faster than absolute view counts can.

Source: backend/agents/scout.py

A video's views divided by the channel's median view count over recent uploads. Surfaces breakout videos against a stable per-channel baseline instead of an absolute view threshold.

Source: backend/services/outlier_detection_service.py

ViralMint classifies outliers in four bands: OUTLIER ≥ 3×, STRONG ≥ 5×, BREAKOUT ≥ 10×, MONSTER ≥ 20×. The 20× tier is the rare viral-template signal worth a response video.

Source: backend/services/outlier_detection_service.py

Composite signal: engagement_rate × 0.40 + recency_bonus × 0.25 + views_score × 0.20 + likes_score × 0.15, multiplied by 100. Weights tuned for short-form discovery, not long-form retention.

Source: backend/services/viral_formula_service.py

A YouTube or TikTok channel where the creator never appears on camera — voiceover, stock or AI b-roll, captions, music. ViralMint's Smart Video pipeline targets this format end-to-end.

Source: /faceless-video-generator/

ViralMint's 11-step auto-pipeline: AI script → voiceover → Whisper-timed captions → Pexels stock or AI clips → background music → finished mp4 with AI-drafted titles, descriptions and tags.

Source: backend/agents/generator_video.py

The opening seconds of a Short that determines retention. ViralMint's analyser extracts the source video's hook structure (statement, question, pattern interrupt) and suggests an angle.

Source: backend/agents/analyzer.py

Parallel trend search across YouTube (50 results), TikTok (30), Douyin (30), Reddit (25) and Google Trends. Fanout via asyncio.gather; per-platform failures isolated so the whole scan continues.

Source: backend/agents/scout.py

Median view count across a channel's last ~15 uploads. Used as the denominator for outlier_ratio so a 1M-subscriber channel and a 1K-subscriber channel are scored on the same axis.

Source: backend/services/outlier_detection_service.py

Advanced SubStation Alpha subtitle format. ViralMint renders viral-style captions by burning ASS — each word fades in at the Whisper-detected timestamp, with three style presets (viral/classic/bold).

Source: backend/services/caption_service.py

OpenAI's speech-to-text model, run locally via the faster-whisper int8 build. Transcribes 10 min of audio in ~45s on mid-range hardware. Used for source transcription + caption timing.

Source: https://github.com/SYSTRAN/faster-whisper

Open-source video extractor supporting 1,800+ sites: YouTube, TikTok, Bilibili, Twitter/X, Reddit, Twitch, Vimeo, Bilibili, Douyin, Instagram, Facebook. The downloader behind every ViralMint scout.

Source: https://github.com/yt-dlp/yt-dlp

Anthropic's standard for tool-augmented AI. ViralMint exposes its full backend (86 tools, 10 modules) as an MCP server so Claude Code, Claude Desktop or Cursor can drive the whole pipeline programmatically.

Source: backend/mcp/

ViralMint's billing model: USD balance topped up via Stripe one-time payments (no subscription, no auto-renew). Each AI action deducts cents from the balance. Daily allowance refills a small starter amount.

Source: cloudbackend/src/config/pricing.ts

Per-user UTC-day spend ceiling (default $30) that limits damage from a compromised token. Resets at midnight UTC. Independent of total balance — protects users without rejecting legitimate big spenders.

Source: cloudbackend/src/utils/balance.ts

AI provider router. ViralMint's cloud backend proxies every AI call (chat, image, video, TTS, music) through OpenRouter — the key never leaves AWS Secrets Manager. Single key, single audit trail.

Source: cloudbackend/src/services/openrouter-service.ts

Pipeline driven by an AI agent rather than a UI flow. ViralMint's planner can chain scout → download → analyse → generate → export in one prompt, with the agent making decisions at each step.

Source: backend/agents/planner.py

GNU Affero General Public License v3. Strong copyleft — derivative SaaS deployments must publish source. ViralMint's OSS variant uses AGPL-3.0 to prevent closed-source forks of the open code.

Source: https://www.gnu.org/licenses/agpl-3.0.html