[Discussion] A new, consolidated maksing workflow (aka "Flexi" masks)

[masterpiga]

TL;DR

Do we want to merge Flexi masks in darktable? If yes, how?

About Flexi

In this post on Pixls I have introduced a working, fully implemented proposal (although still drafty wrt code quality) to revamp the masking UX in darktable.

The new system:

• Uses a much simper, more intuitive mental model
• Is much more powerful and expressive than the classic one
• Subsumes the embedded mask panel + mask editor and provides a unified, streamlined modern masking UI (with margin for improvement, of course)

See the README and user documentation for user-facing docs and more details. I won't repeat the same things here to avoid wasting your time. The Pixls post also has some video demos, even though they are a bit outdated.

The proposal has been received quite enthusiastically by the community, and it has already undergone a few iterations of refinement.

I think that it's a good time to start considering the opportunity of merging it into master.

So, the first question is: do we want to upstream this change?

NOTE: The code is to be perfected, revised and polished (I have been iterating on it heavily and quickly),for sure there are bugs that didn't surface yet, and stuff that can be optimized, but the feature is code complete, it does what it should and it does it pretty well. So, if we decide to upstream it we can start doing it today.

What do you think?

I have not heard a lot from devs on Pixls, so I am really looking forward to hearing your thoughts.

To inform this decision, I (with the help of Claude) have prepared a couple of more technical documents:

• A description of the Flexi data model, and how it diverges from classic. Note that:

  • Flexi doesn't add a new kind of mask, and it doesn't change what a DT_MASKS_GROUP form is
  • All the changes are entirely in what gets read out of bits that are already there, or added to the end of structs that are already there. I.e., it is a relatively small data model change, for a very significant capability and simplicity boost.

• A plan to upstream flexi into darktable codebase.

IMPORTANT: At this stage, the discussion should really not be about what we like or don't like about the UI. If we decide that the change is overall desireable, then we want we will have plenty of time to iterate on the fine print.

@TurboGit of course your opinion is paramount, but I would like to hear from the others too.

Thank you all in advance 😃

[wpferguson]

My first question is what would we lose that we currently have, other than muscle memory :-)

I have not heard a lot from devs on Pixls

I'll speak for myself, and not the rest of the devs. I"ve given up on reading pixls.us conversations. Dealing with all the "opinions" is just exhausting. I still read new issues, tags, and messages but I don't have enough time in my day to wade through all the crap. I much prefer reading the filtered version here without all the argum opini discussion attached. I just looked and it's a 106 entries and still growing. I don't want to spend an hour or two reading through it trying to separate the real information from all the other noise.

I'll also point out that with the advent of AI we now have many more contributors, so the number of PRs and commits has skyrocketed. 5.6.0 was the release with the largest number of commits ever and 5.6.1 is going to add a serious number of commits so the 5.6.x branch will probably have over 2000 commits.

[masterpiga]

Hi Will,

My first question is what would we lose that we currently have, other than muscle memory :-)

I am not sure whether I should read this as an endorsement 🙂

[wpferguson]

I am not sure whether I should read this as an endorsement

It's simply a question. You want to make a change to masking and I'm trying to figure out if something that we currently can do with masking is going to change or go away? And, I don't have hours to read the pixls thread and then try do decipher it.

[TurboGit]

@wpferguson : From what I understand and little tested so far no features are lost. The new approach enable many things currently not possible in an uniform way. And yes you'll have to retrain your brain for the new UI.

[masterpiga]

I'm trying to figure out if something that we currently can do with masking is going to change or go away?

You don't need to read Pixls. If you read the 2nd and 3rd links that I posted above you will get the full picture.

And yes you'll have to retrain your brain for the new UI.

This is actually easier than I thought. After a few hours of testing now it feels strange to go back to the classic masks 🙂

[andriiryzhkov]

So far I noticed that we lost drawn masks creation time controls, which were available in mask manager. That is not critical for majority of masks, but completely wiped controls for AI object mask. Already discussed it with @masterpiga. We will ideate how that could be designed better.

[masterpiga]

@andriiryzhkov I still do not understand what these controls are. Can you show me a screenshot on master?

[andriiryzhkov]

Can you show me a screenshot on master?

Sure - properties in mask manager. Exactly those are available only when AI object mask in creation mode and is not yet converted to vector path.

[masterpiga]

Oh, I see! I used those controls on canvas, but I had never noticed that they had a presence in the mask manager, as I never looked at it while adding a AI mask. I will add them to the UI, thanks!

[TurboGit]

@TurboGit of course your opinion is paramount, but I would like to hear from the others too.

I was notified but never took the time to answer, sorry about that.

The masking support in current Darktable is great, people acknowledge that the masking support we have is amongst the better if not the best. This support has been updated since years, first the parametric masks, then some releases later the drawn masks, a way to combine both and finally the raster masks. Also the mask manager has been created to have a view of the masks being created for an edit.

Fact is, this is working but not in an uniform way. This new PR as I see it (tested once) is to streamline the whole masking support making every kind of mask combinable with others with operators (include, exclude).

To me this move is a great idea and certainly something I'd like to see landed on Darktable.

Now there is quite some risks like breaking complex masking as all masks have to be migrated to the new model. We also want to ensure that creating a simple mask stay simple.

For sure people will have to train again using this new UI/UX, but for a very good reason as I see it today. From the discussion on PixlUS it seems that this new masking support is well received, but I haven't read every messages.

All in all, if this is needed I do bless the work being done here :)

Now, I'll have to test more seriously and report back.

@masterpiga : Thanks for the hard work, another very nice addition to Darktable is cooking :)