llava: computer vision GGUF

  • Clip Handler > mmproj-f16.gguf (624MB)
  • Vision Model > llava-***.gguf (opt anyone you like)
  • provide a link of picture as prompt
  • it will probably give hints to you for complying your text-to-image prompt (reverse engineering) 🎃

how to use

  • compatible with gguf-connector (ggc v) 👻
  • see initial test examples (with just Q2_K - 2.53GB) below

prompt test#1

prompt picture URL: https://raw.githubusercontent.com/calcuis/comfy/master/sd3.5g.png

response: "The image features a computer screen displaying a cartoon of a woman with a fox head, sitting at a dining table. The table is set with a cake, a knife, and a cup, giving the impression that the woman is about to enjoy a meal. The cartoon is displayed in a graphic style, making it visually appealing and engaging."

prompt test#2

prompt picture URL: https://raw.githubusercontent.com/calcuis/suko-puzzle/master/demo.png

response: "The image shows a close-up of a cell phone screen, displaying a game board with a puzzle. The game board is filled with numbers, and there are four
different colored circles surrounding the puzzle. The game appears to be a popular one, as the image is taken from the perspective of someone looking at the screen."

prompt test#3

prompt picture URL: https://raw.githubusercontent.com/calcuis/gguf-connector/master/gguf.gif

response: "The image features a green cartoon character with a mouth wide open, appearing to be yawning. The character has a toothy smile and is wearing a pair of
yellow gloves. The overall scene is a colorful and fun representation of the green character."

feedback from connector: "Awesome!"