You seem to indicate that you didn’t want the thread to deviate from the original topic, but seemingly without much to add to the original topic. These are conversations, and they naturally fluxuate from the original topic. I’m OK with that, as long as it seems like the original topic/question has been discissed/answered.
Yeah I don’t like that idea. People will feel excluded if they’re not within the given time. And closing them manually is just more work.
That is generally for when a topic is too heated, to keep people from emotional replies. I don’t think that’s happened here.
As long as the model is local, AI would be very useful for object detection when creating masks and for tagging images with keywords. Both of these use cases wouldn’t need the (probably heavy) model to be present past tag and mask generation to be useful. The model would have to be well integrated with darktable though. None of the hacky python crap that usually comes with local LLMs.
I’m often quite jealous of the AI denoising that users of proprietary software have access too, but that would certainly need the model to be present for the edit to “keep working”. Even if denoising usually isn’t a creative endeavor (and could be turned off without wreaking havoc), I’m not sure it’s worth it.
It’s about being able to open a raw + xmp in 2, 3, 5 years and have it work and generate the same output as it did when it was originally edited. This means that if darktable deploys these tools as an integrated model, as opposed to an IO model or something like that, it will have to keep all the version of the models it deploys in a repository and provide them alongside darktable.
You could of course, as you imply, apply denoising at export to overcome the need for an ever-present model. But then we’d be working on data at the very last stage of the pipeline. Seems like that could definitely be a recipe for mediocre results.
Exactly. But, preferably in 50 years, not 5 years.
It’s undoubtedly one of the most important aspects of a raw converter to me. Adobe (despite being crap at most other things these days) has actually done a good job handling this in Camera Raw and Lightroom with “Process Versions”.
Camera Raw has been around for 22 years and files using the oldest Process Version should look the same today in the latest Lightroom version. That’s an incredible and important feat.
Thank you for the link. The converter has certainly come along since I removed ACR v5.4 and Elements 6 from my computer long ago and I never did download LightRoom …
AI is demonstrating profound benefits to art and these discussions are becoming academic and we should evaluate the merits of AI modules in darktable when they are developed and not a priori.
Note that while this goal is very noble, it makes refactoring very, very difficult. Increasingly so as time progresses. There are very few examples of software that is
developing rapidly, with amazing additions every 6 months,
compatible with files generated by a decades-old versions of itself (in the sense of not only opening those files, but being able to edit them just you like you did decades ago).
Sorry your comment is non sense. Have you looked at the way we do support old edits? There is no issue for refactoring as this is properly designed. The support is found in a single routine per IOP and has no impact on the rest of the IOP code. So please stop posting wrong information.
But you can’t just keep the old code around without touching it ever after. Look at eg invert (film negatives), it was deprecated five years ago and 20+ updates since.
Yes, I recognize that DT is very well designed. But keeping all old functionality around still has a nonzero cost.
The idea of placing a mask on top of the degraded art work is really clever, I wonder why this wasn’t being done before, albeit manually. Maybe the restoration market has too much influence over these things?
Great use of tech either way, hopefully this catches on and artworks start being preserved in their original state.
That’s a good reason to be selective about adding new functionality. I don’t think it’s a valid reason to stop progressing and adding useful new tools.
I see you meant that we are not removing modules, I was talking about migration code in each IOP.
Indeed we are not removing modules, this will just completely break the edits. At this stage if a module used is not found when starting Darktable it must exit or crash.
