SeedCamp

Cost-optimized AI video generation at scale. Route smart, spend less, ship more.

The Problem

Every AI video tool works great for one video. Try generating 10,000 and you hit the wall: costs explode, quality is inconsistent, failures cascade, and there's no way to prioritize your best products over your long tail.

Companies with 500+ SKUs, vehicle listings, or property portfolios need production-grade orchestration — not a notebook with a for-loop. They need cost control, quality gates, retry logic, and smart routing that sends premium resources where they'll actually drive revenue.

That's the gap SeedCamp fills.

The Solution

Route your top 20% to a premium model. Send the other 80% through a fast model. Save 40%+ on blended cost without sacrificing quality where it matters.

graph LR
    A[Brief] --> B[Script Gen<br/>Seed 1.8]
    B --> C{Smart Router}
    C -->|Hero 20%| D[Seedance Pro<br/>$0.13/video]
    C -->|Catalog 80%| E[Seedance Fast<br/>$0.08/video]
    D --> F[Output + QA]
    E --> F

Loading

The math: At 10,000 items — 2,000 hero × $0.13 + 8,000 catalog × $0.08 = ~$900/year. Blended cost: $0.09/video. Without routing, you'd pay $1,300 running everything on premium.

Quick Start

git clone https://github.com/suboss87/seedcamp.git && cd seedcamp
make install

# Try without an API key (dry-run mode simulates the full pipeline)
DRY_RUN=true make dev    # API on :8000, dashboard on :8501

Set ARK_API_KEY in .env when you're ready for real generation. See .env.example for all options.

Generate your first video:

python3 docs/examples/generate_single_video.py

Try an industry example:

python3 docs/examples/automotive_dealer.py
python3 docs/examples/ecommerce_catalog.py
python3 docs/examples/real_estate_listing.py

5 Patterns You Get

Every pattern is extracted, tested, and ready to reuse in your own AI pipeline.

Pattern	What it does	File	Lines
Tiered Model Routing	Routes requests to optimal model by business value	app/services/model_router.py	~37
Async Task Pipeline	Fire-and-forget generation with polling and timeout	app/services/video_gen.py	~163
Cost Tracking	Per-request token counting and cost attribution	app/services/cost_tracker.py	~81
Batch Orchestration	Semaphore-controlled concurrency with error isolation	app/services/batch_generator.py	~266
Retry with Backoff	Exponential backoff, Retry-After honoring, error classification	app/utils/retry.py	~210

These patterns transfer to any AI workload — image generation, LLM pipelines, audio synthesis. Swap the model call, keep the orchestration.

Adapt for Your Industry

The tier system is a simple enum + route map. Change 3 lines:

# Automotive: certified pre-owned → cinematic walkaround, bulk → quick showcase
class VehicleTier(str, Enum):
    featured = "featured"      # → Seedance Pro
    inventory = "inventory"    # → Seedance Fast

# E-commerce: bestsellers → premium video, long-tail → cost-optimized
class ProductTier(str, Enum):
    hero = "hero"              # → Seedance Pro
    catalog = "catalog"        # → Seedance Fast

# Real estate: luxury listings → cinematic, standard → quick walkthrough
class ListingTier(str, Enum):
    luxury = "luxury"          # → Seedance Pro
    standard = "standard"      # → Seedance Fast

Industry	Hero tier	Catalog tier	Scale
Automotive	Certified, new arrivals	Wholesale, aged stock	300–500K vehicles
E-commerce	Top 20% revenue SKUs	Long-tail catalog	1K–100K SKUs
Real Estate	Luxury listings ($1M+)	Standard listings	500–50K listings
Media	Campaign hero spots	Social cutdowns	100–10K assets

Architecture

Step	What happens	Technology
1. Prompt	Brief + image + tier	FastAPI + Streamlit dashboard
2. Creative	Script or 5-frame storyboard	Seed 1.8 + Seedream/Gemini (plugin)
3. Route	Picks model by business value	Pure function, ~37 lines
4. Generate	Async video + polling	Seedance Pro / Fast via ModelArk
5. Validate	Quality scoring + safety eval + human approval	6 dimensions, configurable thresholds

Image providers are plugins — Seedream and Gemini included, add your own by implementing one file.

Deploy Anywhere

Platform	Guide	Setup
Local	make dev	No Docker needed
Docker	deploy/docker/	make docker-up
GCP Cloud Run	deploy/gcp/	Terraform
AWS ECS Fargate	deploy/aws/	Terraform
BytePlus VKE	deploy/byteplus/	K8s manifests
Railway	deploy/railway/	One-click
Render	deploy/render/	One-click

Links

• Quick Start Guide — Railway, Render, Docker in 30 min
• Deployment Guide — All platforms, step by step
• API Docs — Swagger UI (run locally)
• Market Research — Data behind the positioning
• Contributing — Good first issues included
• Security — Vulnerability reporting

---

Built by Subash Natarajan · Powered by BytePlus ModelArk