Edify Image
Image GenerationNVIDIA · Picasso · v1.0 · released
About
Edify Image is NVIDIA's enterprise text-to-image model — trained exclusively on commercially-licensed data through the Picasso platform. Positioned for advertising, design, and content-creation workflows where copyright-clean image generation is a hard requirement.
Intended use: Commercial image generation for marketing, advertising, product photography, and creative tooling where IP-cleared training data is required. Fine-tunable on customer corpora.
Architecture
- Type
- diffusion
- Parameters
- 4.0B
Cascaded latent diffusion text-to-image model — T5-style text encoder + base 1024×1024 latent diffusion U-Net + upsampling diffusion stages. Trained on NVIDIA's curated licensed-content corpus (Getty Images, Shutterstock partner data). Available exclusively as a hosted API via NVIDIA Picasso, with optional fine-tuning on customer-licensed imagery. Inference on H100 / L40S with TensorRT optimization.
Memory
- Weights (BF16)
- 8.00 GB
- Weights (FP8)
- 4.00 GB
- Activation estimate
- 4.00 GB
Pricing
$0.040 per image (1024x1024)
Provenance
- Source
- developer.nvidia.com
- License
- nvidia-picasso-commercial-license
- Last verified
- 2026-06-25