QWEN3-4B-VL Caption GUI

• This model has been finetuned with image and text pairs at 1024px and shows high affinity with limited hallucination on NSFW task.

• Full FP32 Training with blockwise conversion to FP8 (Key Blocks in FP32)

• This model has limited video caption ability.

This model has been heavily tested and excels at most NSFW and SFW captioning task.