FastVLM-7B ComfyUI Node

A custom ComfyUI node for Apple’s FastVLM-7B vision-language model.
This node lets you pass an image + instruction and returns a generated text response. Please check Apple License before downloading the Model

✨ Features

• 🔹 Uses apple/FastVLM-7B from Hugging Face.
• 🔹 Accepts ComfyUI images (B,H,W,C float tensors).
• 🔹 Converts to PIL image internally for model input.
• 🔹 Text input (instruction) with multiline support.
• 🔹 Output is a STRING containing the model’s answer.
• 🔹 Automatically checks ComfyUI/models/LLM/FastVLM:
  • Creates folder if missing.
  • Downloads model into it on first run.
• 🔹 Works on GPU (fp16) or CPU (fp32) depending on your system.

---

support us on Patreon

Check our TBG Enhanced upscaler and Refiner for ComfyUI at https://github.com/Ltamann/ComfyUI-TBG-ETUR