Creators never integrate a model provider. They pick from murl’s model registry, one catalog that carries everything the studio, the apps, and the authoring agent need to use a model well.
What the registry knows per model
- Posted pricing: per-unit prices (per run, per second, per megapixel) that feed the credit price shown before every run, kept in sync with provider cost by a scheduled pricing job.
- Capability tags: text-to-image, image-to-video, start/end-frame, reference images, LoRA support, lip-sync audio/video sockets. The studio uses these to know which blocks can chain into which.
- Guidance cards: researched, per-model briefs (what it’s best for, style tags, prompting advice). These are rendered into the authoring agent’s context, so a brief like “cinematic product shot” lands on the right video model rather than a name match.
- Provider bindings: a model can have multiple upstream providers with health tracking and price-aware ordering; calls fail over automatically. Providers are platform-internal and never surface to creators or apps, which see model display names only.
Coverage
The library spans 75+ models: state-of-the-art image generation and editing, a deep video bench (text-to-video, image-to-video, start/end-frame control, lip-sync, add-sound), and text models for prompt work and in-app logic. All calls route through one AI gateway for observability and failover.
One registry, three consumers
| Consumer | What it gets |
|---|---|
| Studio blocks | typed sockets, chaining rules, posted per-block prices |
| Apps (SDK) | estimateCost / callModel by model id, registry-validated chains |
| Authoring agent | guidance cards + capability filters for brief-driven model picks |
Adding a model to the registry lights it up across all three, with no per-surface integration.