Some feature advices

[925-Studio]

Thank you for making this model and nodes, it works straight forward and almost great. But I still have some advices if I may.

1. Add 'Keep model loaded' feature like Florence 2 does. It will save huge amount of VRAM for further generation (clean the VRAM after captioning). I can only use 4 bit on my 4080S which has 16GB VRAM and the image generation speed would be 2x slower after using your caption method.
2. Add 'Seed' option.
3. Add 'None' for extra options.
4. Sometimes it generates watermarks info even there's no watermark on the image. Also I can't find the option to ignore it.

I know it's a lot asks and I do hope this model can go further and further. Anyway, thanks again for your hard work.

[fpgaminer]

There is a "None" for extra options, just ComfyUI doesn't show it. Wasn't sure the best way to fix that. But if you select the top option and then arrow to the left it'll go blank.

I'll take a look at 1 & 2.

[925-Studio]

There is a "None" for extra options, just ComfyUI doesn't show it. Wasn't sure the best way to fix that. But if you select the top option and then arrow to the left it'll go blank.

I'll take a look at 1 & 2.

Thank you for your reply. Looking forward to those.

[Noatitauaggi]

There is a "None" for extra options, just ComfyUI doesn't show it. Wasn't sure the best way to fix that. But if you select the top option and then arrow to the left it'll go blank.

I'll take a look at 1 & 2.

And why so few extra_options?

[Slartibart23]

is it possible to edit the nodes.py. there are rules for captioning. if i change them, cumfyui can no longer find the nodes :/
I would like to strengthen the recognition of hairstyles, suppress the mention of watermarks and maybe one or two other things ;-)