comfyuiblog/BitDance-14B-64x-fp8-comfyui

BitDance 14B (FP8 & FP16) for ComfyUI: Low VRAM Optimization

This repository contains the optimized FP8 and FP16 model files required to run the BitDance 14B model locally inside ComfyUI on consumer GPUs (12GB - 24GB VRAM).

These weights have been manually quantized, tested, and verified to prevent CUDA Out of Memory (OOM) errors and "Black Screen" crashes during video generation.

File Placement Instructions:

• BitDance_14B_MainModel_FP8.safetensors: The main diffusion model. Place this in your ComfyUI/models/diffusion_models/ folder.
• BitDance_TextEncoder_FP8.safetensors: The required text encoder. Place this in your ComfyUI/models/text_encoders/ folder.
• BitDance_VAE_FP16.safetensors: The VAE for decoding. Place this in your ComfyUI/models/vae/ folder.

---

Required Custom Node & Workflow (.json)

To successfully run these models without node errors, you must install the custom BitDance node and use the optimized node routing for low VRAM.

• 💻 Install the Custom Node: ComfyUI-BitDance on GitHub
• 📖 Download the exact .json workflow and step-by-step guide: 👉 How to Run BitDance 14B in ComfyUI (Low VRAM Workflow Fix)
• 📖 Youtube Tutorial Guide: 👉 https://www.youtube.com/watch?v=4O9ATPbeQyg

---

About the Creator

Tested, compiled, and maintained by Esha Sharma, Founder of AI Study Now. I build and document custom ComfyUI workflows, GGUF optimizations, and local AI solutions for consumer hardware.

• 📺 Watch the video tutorial: @ComfyUIworkflows
• 💬 Follow for updates: X/Twitter