Stable Diffusion 2.1

The Stable Diffusion v2-1 is a diffusion-based text-to-image generation model utilizing a pretrained OpenCLIP-ViT/H text encoder. It is a Latent Diffusion Model (LDM) fine-tuned from the Stable Diffusion v2 checkpoint. The model underwent additional training:

• 55k steps on a dataset filtered with LAION's NSFW detector (p_unsafe=0.1)
• 155k further steps with p_unsafe=0.98.

The model employs an autoencoder to encode images into latent representations, followed by a UNet backbone trained with a reconstruction objective. It uses the v-objective for diffusion noise prediction. Trained on 32×8×A100 GPUs with AdamW optimizer (learning rate: 0.0001). Generates high-resolution images (up to 768x768) and supports text-guided generation.

Announce Date: 06.12.2022
Parameters: 3.5B
Context: 77
VRAM requirements: 1.6 GB using 4 bits quantization
Developer: Stability AI
License: OpenRail++

Use our pre-built public endpoints to test inference and explore Stable Diffusion 2.1 capabilities.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.