Model Card
NVIDIA SANA is a text-to-image diffusion model built by NVIDIA Research with MIT's Han Lab and Tsinghua University, designed for efficient high-resolution image synthesis.
It replaces the standard attention layers in a diffusion transformer with linear attention and pairs that with a deep-compression autoencoder that reduces images to 32x fewer latent tokens than the 8x compression typical of earlier models. It also uses a small decoder-only LLM as its text encoder instead of T5. Together these choices let it generate images up to 4096x4096 without the usual quadratic cost of scaling resolution.
NVIDIA's SANA paper reports its 0.6B-parameter version is competitive in quality with FLUX-12B while being 20 times smaller and over 100 times faster in measured throughput. That combination of speed and Apache 2.0 licensing suits workloads that call for high-resolution output or high generation volume through an API, where per-image latency and cost add up quickly.
Cost Per Image $0.0015
per generation
Configuration second
resolution
Release Date Oct 14, 2024
Code Example
Use SANA in your app with the Puter.js AI API — no API keys or setup required.
// npm install @heyputer/puter.js
import { puter } from '@heyputer/puter.js';
puter.ai.txt2img("A serene mountain landscape at sunset", {
model: "nvidia/sana"
}).then(image => {
document.body.appendChild(image);
});
<html>
<body>
<script src="https://js.puter.com/v2/"></script>
<script>
puter.ai.txt2img("A serene mountain landscape at sunset", {
model: "nvidia/sana"
}).then(image => {
document.body.appendChild(image);
});
</script>
</body>
</html>
More AI Models From NVIDIA
Nemotron 3.5 Lightning
Nemotron 3.5 Lightning is a 30B-parameter Mixture-of-Experts model from NVIDIA with 3B active parameters per token, using a hybrid Mamba-2, MoE, and attention layer architecture. It is designed for high-volume, low-latency execution inside multi-agent systems, where a larger reasoning model like Nemotron 3 Ultra plans and delegates, and Lightning handles tool calls, code review, and other repetitive subagent tasks. NVIDIA reports up to 4x higher output throughput than similarly sized open models and, on its PinchBench agent benchmark, 86% accuracy while completing 10,000 tasks 30% faster than Qwen3.6 35B at similar accuracy. The model ships with built-in multi-token prediction and draft models for speculative decoding, and is available in NVFP4 and BF16 checkpoints under the permissive OpenMDW-1.1 license with open weights and training recipes. Choose it for agent harnesses handling frequent, narrow calls, tool validation, or subagent delegation where response speed matters more than broad reasoning depth.
ChatNemotron 3 Ultra 550B A55B
Nemotron 3 Ultra 550B A55B is NVIDIA's open-weight frontier reasoning model with 550B total and 55B active parameters, built on a hybrid Mamba-Transformer Mixture-of-Experts architecture. It supports a 1M token context window and is designed for long-running agentic workflows, complex multi-step reasoning, and high-accuracy tasks across code, math, and science. NVIDIA reports up to 5.9x higher inference throughput than comparable open MoE models. On the Artificial Analysis Intelligence Index it scores 48, leading US open-weight models and delivering the highest non-hallucination score in its comparison set (78.7 on AA-Omniscience). Choose it for production agentic pipelines, deep document analysis, or reasoning-heavy API workloads where both accuracy and throughput matter.
ChatNemotron 3.5 Content Safety
Nemotron 3.5 Content Safety is a compact 4B-parameter multimodal guardrail model from NVIDIA, fine-tuned from Google Gemma-3-4B, designed to moderate both inputs and outputs of LLMs and VLMs. It classifies prompts and responses as safe or unsafe across 23 safety categories based on the Aegis v2 taxonomy, supports 12 languages, and accepts both text and image input. An optional reasoning mode provides step-by-step chain-of-thought traces explaining each decision — useful for auditing and policy tuning. Despite its 4B size, it leads external multimodal safety benchmarks including the top harmful-F1 score on VLGuard, matching or beating 8–12B models. It also supports custom operator-defined content policies enforced at inference time. Choose it for prompt and response moderation pipelines, safety evaluation of LLM outputs, or as an inference-time guardrail in enterprise AI applications requiring explainable, policy-aware content filtering.
Frequently Asked Questions
You can access SANA by NVIDIA through Puter.js AI API. Include the library in your web app or Node.js project and start making calls with just a few lines of JavaScript — no backend and no configuration required.
SANA is free to integrate using the Puter.js AI API. With the User-Pays Model, you can add AI to your app for $0, since users cover their own AI usage through their Puter account.
| Price | |
|---|---|
| Per image | $0.0015 |
SANA was created by NVIDIA and released on Oct 14, 2024.
Yes — the SANA API works with any JavaScript framework, Node.js, or plain HTML through Puter.js. Just include the library and start building. See the documentation for more details.
Add SANA to your app for free
Developers can integrate SANA for free using the Puter.js AI API.
With the User-Pays Model, each user covers their own AI usage instead of the developer.