<h2 id="qwen3-8b-imagevideo-caption-(uncensored)">QWEN3-8B Image/Video Caption (Uncensored)</h2><a target="_blank" rel="ugc" href="https://civarchive.com/articles/27449/qwen-videoimage-caption-full-nsfw-standalone">Version 2</a> - This version is highly attuned to NSFW content. However do to image only training it may generate some video captions as image.<ul><li>This version requires 24GB or more of VRAM</li><li>Full Finetune (NOT A LORA MERGE) of the 8B parameter model (Vision Frozen)</li><li>BF16/TF32 training unfortunately do to the size of the model Adam8bit needed to be used.</li><li>Version 2 Can use nearly any LLM prompt - Version 1 should use the prompt given in whole or in part.</li><li>Details regarding training of version 1 can be read about <a target="_blank" rel="ugc" href="https://civarchive.com/articles/27203">here</a>.</li></ul><hr />Note: No image size safety is built in I have captioned 4k images which will be processed to a very large tensor shape - however reduction to 1k images is recommendI have an Ampere series card and can not convert this to FP8 or NF4 in high quality. If you have experience converting models with Linux and Transformer Engine DM me.

Files with SHA256: f2dea1b89c9953455b79f4b13ebf3f7ac0833b0658228922c7c2ad843beb2826