Backpropagating through Qwen3-VL-8B's vision stack to the input image Researcher @_smitop demonstrated backpropagating through the Qwen3-VL-8B vision stack to the input image, optimizing the image to resemble a target image, and released the work as a Hugging Face Space called "Qwen3-VL Soft-Token Inversion" by syvb. The post, dated Nov 18, 2026, credits inspiration to @voooooogel's Nov 16 work on deep-dreamed images that activate neurons and chain lens readouts in Qwen 3.6 27B. loops @ smitopbackproping through the Qwen3-VL-8B vision stack to the input image, to optimize the input image to make it like look like a target image00:001:09 AM · Nov 18, 2026·565Views131 loops @ smitopbackproping through the Qwen3-VL-8B vision stack to the input image, to optimize the input image to make it like look like a target image00:001:09 AM · Nov 18, 2026·565Views131 loops @ smitop14hQwen3-VL Soft-Token Inversion - a Hugging Face Space by syvbFrom huggingface.co22103 loops @ smitop14h kinda inspired by x.com/voooooogel/sta… thebes@voooooogelNov 16deep dreamed images that activate neurons and chain lens readouts in qwen 3.6 27b42