Abiray/Anima-base-v1.0-FP8-INT8

Anima-base-v1.0-FP8

This repository contains FP8 (E4M3FN) and INT8rowwise weights for Anima-base-v1.0.

Anima is a 2B parameter diffusion model built on the NVIDIA Cosmos architecture. This FP8 version provides a significant reduction in VRAM usage while maintaining near-original fidelity.

🚀 Model Details

• Architecture: Cosmos (Diffusion)
• Precision: FP8 (E4M3FN), INT8rowwise
• Parameters: 2 Billion
• Original Model: circlestone-labs/Anima

🧩 Required Components

You will need to download the following files from the original repository to complete the pipeline:

1. Text Encoder: qwen_3_06b_base.safetensors
2. VAE: qwen_image_vae.safetensors

🎮 How to Use in ComfyUI

1. Place the anima-base-v1.0-fp8.safetensors file in your ComfyUI/models/checkpoints/ or unet/ folder.
2. Use the Load Checkpoint or UNET Loader node.
3. In the Checkpoints or Model Sampling settings, ensure your weight_dtype is set to fp8_e4m3fn.

⚖️ License

This model is subject to the NVIDIA Cosmos License. Users must adhere to the terms specified by NVIDIA and Circlestone Labs.