Mohammad Norouzi#763
Added an Ideogram 4 auto captioner to AI Toolkit. It automatically does the boxes and the json for you. I tested with Qwen-3-VL 8B and it works quite well. I even added a little toggle to view the boxes in the dataset viewer.
Users appreciate the Ostris AI Toolkit's Ideogram 4 Auto Captioner addition because it delivers practical time savings and quality-of-life features like model toggles.
@ostrisai Is it using the full weight encoder? Or the fp8 scaled? Inquiring for vram limit purposes lol
@nonRealBrandon @PhotogenicWeekE Yeah, I will probably expose the whole thing, but it is long. I only expose part of it.
@ostrisai Tiny request, would it be possible to add another toggle for "duplicate this dataset and rewrite the prompts"?
@ostrisai @PhotogenicWeekE Would be nice if the caption prompt was exposed and had a larger text field
@ostrisai huh, that Qwen-3-VL model actually pulling its weight on auto captioning is nice to see
toggle in the viewer is the real QOL flex
@ostrisai Thank you this is much needed
@ostrisai Thanks, that legit just saved me a lot of time making my own!
@ostrisai Cool, helpful for data labeling but seen it all before.
@el_mejnun I ran into this issue too. I can make a separate extension so it has both on the same dataset. I plan on doing that soon.
@rugbist_ @ostrisai I think there's a good chance it's what they used during training... would make sense as they're using it for the model's encoder...