ApX logoApX logo

Stable Diffusion 3.5 Large Turbo

Parameters

8.1B

Modality

Image

Auxiliary Parameters

6.16B

Architecture

Diffusion Transformer

License

Stability AI Community License

Release Date

22 Oct 2024

API Pricing (per 1M)

Self-hosted only

Hardware Sizing

Precision

Workload Memory: 6.64 GB

INT4 · 1024x1024

DiT Backbone Weights: 3.77 GB

VAE Decode Workspace: 0.37 GB

Runtime & Graphs: 2.50 GB

Minimum Required Hardware

1x NVIDIA L4 24GB

6.6 / 21.6 GB usable

1x RTX 3060 12GB

6.6 / 10.8 GB usable

Apple M4 24GB

6.6 / 18.0 GB usable

Determine requirements for your specific workload:

Open in VRAM Calculator

Architecture Diagram

Noisy Image LatentsLatent Image Space (VAE)Patch Embedding2D Latent PatchifyHidden: 4.1k · Text: 6.16B enc · Timestep (t)x 38 DiT blocksAdaLN ModulationNorm · Timestep scale/shiftJoint / Cross-Attention38 headsHead dim: 64+AdaLN ModulationNorm · Timestep scale/shiftFeed-Forward (MLP)Activation+Final Norm (AdaLN Scale)Predicted Noise / LatentsDenoised Latents to VAE

About Stable Diffusion 3.5 Large Turbo

Stable Diffusion 3.5 Large Turbo is an adversarial diffusion-distilled variant of SD 3.5 Large, capable of generating detailed photorealistic images in only 4 sampling steps without quality degradation.

Technical Specifications

Attention

Attention Heads

38

Attention Head Dimension

64

Dimensions

Auxiliary Parameters

6.16B

Hidden Dimension Size

4,096

Number of Layers

38

About Stable Diffusion 3.5


Other Stable Diffusion 3.5 Models