gen-floor — $/image and $/second floor for generative-media APIs

Cheapest model right now, per category. Updated every 15 minutes from fal.ai and hourly from Replicate. Per-megapixel and per-clip prices are normalized to 1 MP / 5 s where exact equivalents aren't published — see About for the assumptions.

Biggest price drops and spikes vs the snapshot taken closest to days ago. Only models whose pricing unit didn't change are included.

Drops

Spikes

Model	Lab	Category	Source	Price	Unit	Per image	Per second	History

Last 50 cron runs. Each row records seen/priced counts and any error message — useful for catching schema drift in upstream sources.

Started	Source	Status	Seen	Priced	Duration	Error

About gen-floor

gen-floor is a live floor-price tracker for generative-AI media APIs. Token-floor for pixels and frames. It answers: which provider is cheapest right now for text-to-image / text-to-video / image-edit, which model just dropped this week, and what's the per-model price history.

Data sources

fal.ai catalog — GET https://fal.ai/api/models?page=N&size=40 — public JSON, ~1,350 models, refreshed every 15 minutes. We parse the pricingInfoOverride markdown for per-image, per-megapixel, per-second, and per-video prices.
Replicate model pages — GET https://replicate.com/<owner>/<slug> — server-rendered HTML, curated slug list (~30 high-signal models), refreshed every 60 minutes. Pricing is parsed from natural-language phrases like "333 images for $1" or "$0.40 per second".

Normalization assumptions

per_megapixel → treated as equivalent to per_image at 1 MP. Most demo configurations on fal.ai and Replicate default close to 1 MP.
per_video → divided by 5 to derive an approximate per_second figure (5 s is the modal default-clip length across Kling, Veo, Pixverse, Wan).
Drop / spike percentages compare the latest snapshot to the snapshot closest to N days ago, but only when the pricing unit hasn't changed between snapshots.

What's not here

Hardware-second pricing (e.g. fal.ai's $1.89/h H100), per-token text-mode pricing, and training prices are intentionally excluded — those are not comparable units.

No mock data

Every row is fetched live. If both sources fail, the dashboard shows "awaiting data" rather than seeding fake numbers. The full per-run log is at the Fetch log tab.

JSON endpoints

GET /gen-floor/health
GET /gen-floor/api/floors
GET /gen-floor/api/models[?category=&source=&q=]
GET /gen-floor/api/model/<id>/history
GET /gen-floor/api/movers[?days=7&limit=10]
GET /gen-floor/api/categories
GET /gen-floor/api/fetch-log[?limit=50]