latent-diffusion-f8-large-safetensors

Latent Diffusion f8-large (Converted to SafeTensors)

This is the historic original Latent Diffusion Model (LDM) text-to-image checkpoint (txt2img-f8-large) from CompVis, converted into the safe and fast .safetensors format.

⚠️ Important Notice: Custom Nodes Required

This model utilizes a non-standard architecture (including a BERT-based text encoder with dim: 1280 and vocab_size: 30522 alongside an f8 autoencoder).

• Standard nodes will not work: Default loaders in ComfyUI or other interfaces will throw compatibility/structure errors if you try to load this file directly.
• Requirements: You must use the dedicated custom nodes specifically designed to initialize and run this Latent Diffusion pipeline.

Model Details

• Developed by: CompVis (LMU Munich)
• Architecture: Latent Diffusion v1 (with BERT Text Encoder)
• Text Encoder Dimensions: dim: 1280, vocab_size: 30522
• Format: SafeTensors

Internal Metadata

The .safetensors file contains embedded string metadata specifying the target architecture dimensions, allowing custom loaders to automatically parse its internal structure without relying on external .yaml configuration files.