# Models on Nesa: Playground and Custom Uploads
## Model Playground & Custom Models
The **Model Playground** is where developers explore, evaluate, and deploy models on Nesa. It combines an extensive on-chain model catalog with the ability to upload and operate custom models directly on the network.
> **Note:** Custom model uploads are currently available to **Nesa Pro users only**. Pro users may have up to **5 custom models** active on the network at any given time.
***
### Model Playground
Nesa maintains a large and continuously growing catalog of on-chain models spanning text, image, video, audio, 3D, bio, and multi-modal tasks.
The Playground allows builders to:
* Browse and compare models by modality and task
* Inspect performance signals such as latency and usage
* Select models for use in DAIs or downstream applications
All models in the Playground run on Nesa’s decentralized inference infrastructure, providing consistent execution and observable performance.
***
### Extensive Model Support
The Model Playground provides access to a **large, heterogeneous catalog of production-ready models**, spanning language, vision, audio, and video workloads. Models are exposed as first-class on-chain assets, each with clear metadata, performance signals, and usage characteristics.
Out of the box, the Playground includes:
* **Text and reasoning models**\
Instruction-tuned and reasoning-focused LLMs (e.g., Llama, Qwen, DeepSeek, GPT-family variants) optimized for dialogue, coding, retrieval, and multi-step reasoning, with visible latency and usage statistics.
* **Image generation models**\
Diffusion and DiT-based models for artistic, photorealistic, and design-oriented generation (e.g., anime-style illustration, typography-aware posters, fast preview models), supporting a wide range of resolutions and inference speeds.
* **Video generation models**\
Short- and medium-horizon video models designed for cinematic content, storytelling, and animation, with different trade-offs between quality, latency, and clip length.
* **Audio and speech models**\
Models for speech synthesis, music generation, and audio understanding, exposed with consistent inference and deployment interfaces.
* **Multi-modal and domain-specific models**\
Models that combine text, vision, or audio inputs, as well as specialized models tuned for branding, design, reasoning, or creative workflows.
Each model card surfaces **practical signals** such as:
* Average inference latency
* Community usage and ratings
* Supported modality and task
* Selection and deployment status
This allows teams to **compare, prototype, and deploy models quickly** without managing hosting, scaling, or execution pipelines themselves, while retaining the option to upload and operate custom models when needed.
***
### Uploading Custom Models
In addition to the public catalog, Nesa allows developers to **upload custom models** and run them on the network.
Custom models can be registered through the **Build on Nesa** flow and configured using:
* A Docker image or GitHub repository
* Model metadata and description
* Modality and task definitions
* Visibility and access settings
Once uploaded, custom models become first-class citizens on the network and can be used by DAIs or other applications.
> **Pro access:** Custom model uploads require a **Nesa Pro** account. Each Pro user may register up to **5 active custom models** at a time.
***
### Consistent Execution and Presentation
Uploaded models follow Nesa’s standardized execution and presentation pipeline. This ensures:
* Consistent inference behavior across models
* Unified UI and metadata for discovery
* Compatibility with DAIs, staking, and incentives
Optional features such as encryption, visibility controls, and future access policies can be configured per model.
***
### Summary
The Model Playground gives builders immediate access to a broad ecosystem of on-chain models, while custom model upload enables teams to deploy proprietary or specialized models without operating their own infrastructure.
Together, these capabilities make Nesa a practical environment for building, testing, and scaling decentralized AI applications.
