Phase 0: Generative Media Fundamentals · 45 min · Python
Economics & Cost Models
If you can't calculate the cost per generation, you can't ship generative media to production.
Hiring signal: Cost modeling is a first-class engineering concern in generative media — interviewers test whether you can reason about per-generation economics and GPU-hour costs.
What you will learn
- Calculate per-generation costs: per-image ($0.01–0.10), per-video ($0.20–0.50), per-song ($0.05–0.50)
- Reason about GPU-hour costs: $2.00 (H100) to $5.00 (B200) and how they translate to per-generation economics
- Apply the speed-quality-cost trade-off triangle to make deployment decisions
- Project monthly costs for a production generative media service at different scales
The Problem
A startup launches a generative media service. They price it at $0.05/image. After three months, they're losing money — each image costs $0.08 in API calls because they're using FLUX Pro instead of Schnell, they generate 8 candidates per request (generate-N-pick-best), and they upscale every output. They didn't model the cost per generation before pricing.
Cost is not an afterthought in generative media — it's a first-class engineering concern that drives every decision: which model, which deployment strategy, how many candidates to generate, what quality threshold to set.
What you'll build
A cost calculator that takes a usage profile (N images/day, M videos/day, P songs/day) and computes daily/monthly costs across deployment strategies. Includes break-even analysis for self-hosting vs API.
Per-Generation Costs
The fundamental unit of generative media economics is the cost per generation.
Image Generation
| Model | Provider | Cost/Image | Speed | Quality |
|---|
| FLUX.1 Schnell | fal.ai | $0.003 | ~2s | Good (4 steps) |
| FLUX.1 Dev | fal.ai | $0.025 | ~8s | High (20 steps) |
| FLUX.1 Pro | fal.ai | $0.05 | ~12s | Highest |
| SDXL (self-hosted) | RunPod H100 | ~$0.01* | ~5s | High (with LoRAs) |
| Midjourney V8 | Midjourney API | $0.10 | ~15s | Best aesthetics |
| DALL-E 3 | OpenAI API | $0.04 | ~10s | Good |
*Self-hosted cost = GPU-hour cost ÷ images per hour. H100 at $2.00/hr generating ~200 images/hr = $0.01/image.
Video Generation
| Model | Provider | Cost/Video | Duration | Resolution |
|---|
| Sora 2 | OpenAI API | $0.30-0.50 | 20s | 1080p |
| Veo 3.1 | Google API | $0.20-0.50 | 10s | 4K |
| Kling v2.6 | fal.ai | $0.25-0.40 | 10s | 4K 60fps |
| Runway Gen-4 | Runway API | $0.30-0.50 | 10s | 1080p |
| Pika 2.0 | Pika API | $0.08 | 5s | 1080p |
Audio/Music Generation
| Model | Provider | Cost/Song | Duration | Notes |
|---|
| Suno v5.5 | Suno API | $0.10 | 3min | Full songs with vocals |
| Stable Audio 3 | Stability API | $0.05 | 3min | Open weights option |
| MusicGen Large | Self-hosted | ~$0.02* | 30s | Open source |
| ElevenLabs Music | ElevenLabs | $0.15 | 2min | Commercial licensing |
3D Generation
| Model | Provider | Cost/Model | Format | Notes |
|---|
| Tripo AI | Tripo API | $0.15 | GLB | Game-ready topology |
| Rodin AI | Hyper3D API | $0.25 | GLB/FBX | Photorealistic |
| TRELLIS 2 | Self-hosted | ~$0.05* | OBJ/GLB | Open source, best quality |
Unlock the full lesson
You've read the first 2 sections. The rest of this lesson covers GPU-Hour Costs, The Speed-Quality-Cost Triangle, Pipeline Cost Stacking, Monthly Cost Projections, Key Takeaways, What's Next — plus a hands-on lab, quiz, and project artifact.
Create a free account to unlock Phase 0 and Phase 1 of every course — no credit card.
Browse all courses · View pricing · DeVenture Academy