The Base Models page is Layer's catalog of every model we currently support — across image, video, 3D, audio, and text. It's where you can see exactly what's powering your generations, how each model is rated, what each one is best at, and which models are turned on for your workspace.
📘 Base models aren't your custom-trained models. These are the foundation models from providers like OpenAI, Google, Black Forest Labs, and ByteDance. They power everything in Layer — including the custom styles and models you train on top of them, plus every other model-driven feature.
Finding a model
Base Models lists every model available in your workspace, ordered by rating so the strongest options sit at the top. Three controls help you narrow the list:
Search — type a model's name to jump straight to it
Modality filter — show only Image, Video, 3D, Audio, Text, or Playable models (or All)
Filter by capabilities — narrow to models that support a specific capability, like transparency or image editing
What each model shows
Open any model and you'll see:
Provider & modality — who built it (e.g. OpenAI) and what it generates (Image, Video, …); standout models are tagged Featured
ELO rating — a quality score from the Artificial Analysis leaderboard (e.g. GPT Image 2 = 1339). Higher means stronger in head-to-head comparisons. (Text models show an Intelligence Index instead.)
Popularity — how widely the model is being used
Timing & Price — typical generation time (2–4m) and cost in Creative Units (e.g. 6 CU per megapixel for images; text models are priced per generation)
Capabilities — what the model supports (e.g. image-to-image, transparency) — this varies model to model
Best for — a quick list of where it shines
Choosing which models are available in your workspace
If you're a workspace admin, each model has an Enable / Disable toggle in the top-right of its page. Use it to control which models your workspace can generate with — turn off the ones you don't want in play, and keep the lineup focused on what your team actually uses. A disabled model stays in the catalog (so you can read about it and turn it back on anytime), it's just not available for new generations.
See a model's supported sizes
To see the resolutions and aspect ratios a model supports, open the Try it panel on the model's page and click the size selector. You'll see exactly what's available — and it's model-specific. GPT Image 2, for example, supports 1K, 2K, and 4K, each with its own set of aspect ratios (3:2, 1:1, 2:3 at 1K, up to 16:9 / 9:16 at 4K). The selector even shows the exact pixel dimensions and megapixels for the size you pick:
💡 A model's page is also a great way to learn it: browse "Made with this model" for real examples (and reuse their exact settings with Use parameters), or use Try it to generate right there without leaving the page.
How models are rated — and why you can trust the lineup
Layer doesn't pick models by hunch. The ELO rating comes from Artificial Analysis, an independent benchmark that ranks models by head-to-head results — so the scores reflect real, comparative performance, not marketing claims. Paired with the popularity figure, you can see at a glance which models are both proven and widely used. Every model on the page has earned its place.
You don't have to choose — the Agent does
Here's the payoff: because Layer knows each model's ratings, capabilities, and supported sizes, the Agent automatically picks the best model for your request. Ask for a stylized game icon and it reaches for a strong image model; ask for a short clip and it chooses a capable video model.
And if you have a preference, just ask for it by name — "use GPT Image 2," "make this video with Kling" — and the Agent will use that model. You can also star the models you rely on for quick access from the sidebar.
What's next
Make something → [Generate your first asset]
Steer the Agent → [Writing better prompts]
Drive a model yourself → [Generate and edit assets directly]