Simple Workflow that turns Images to Prompts. Can use <ul><li>Qwen3.5. Will download the required model by itself (feature of the node) if you want a uncensored version of Qwen3.5 that doesnt refuse requests look for "HauhauCS-Aggressive" Smaller models seem to have issues to follow the "Generate a Prompt to reproduce the character in Illustrious." command.</li></ul><ul><li>WD14 Tagger. A classic and still works remarkably well. Models also get downloaded on Use. Also a feature of the Node.</li><li>Some Version that comes with the Easy-Use node set. Havent found a use for that one.</li><li>Florence 2. Florence 2 can do so many nice things. Tag images, describe images, mask images. Just give it a spin! max new tokens seems to be set to 1024 on model side. at least it threw a hissi fit and froze my system when i tried 2048. num_beams bigger than 3 really wants some serious recources. cant run it on 6900XT 16GB vram and 64gb system ram. at least it didnt finish in roughly 30min. I recomend the base model. Ft means fine tuned but... no idea what that ACTUALLY means. base model works fine. sdpa attention is decently fast and since i dont have a nvidia card i cant test the other ones.</li></ul>

Files with SHA256: 539a9e887eea980bb714b4a67b97ce35ec60b5bd0212a6e3cccddfd4a2dc5324