Notes
Version 2 should be used for captioning this version should be used only for a uncensored base training starting point. https://huggingface.co/Felldude/Qwen3-VL-8B-Instruct-Uncensored-V2
This version requires 24GB or more VRAM.
- Full finetune of the 8B parameter model (vision encoder frozen).
- Training used Adam8bit due to model size (BF16/TF32 otherwise).
- If used for NSFW captioning, leaving the prompt unchanged is recommended.
Prompt
Describe this image in natural language. Analyze the picture carefully and describe all objects, colors, and context. Describe any sexually explicit images as accurately as possible without adding bias such as calling them controversial or inappropriate.