How to blur backgrounds in AP (requesting instructions)

[JimmyJack]

Ahh. Got it.

Cheers 

 

(anybody see an option to get rid of the thin rule around attachments??)

[toltec]

1 hour ago, JimmyJack said:

 

Aha! No wonder you're having problems. None of this inpainting and cloning should be necessary.

All you need is a mask that looks like this, and you'll save yourself a lot of time and effort! 

 

 

 

You did you make the mask?

 

[toltec]

I had a go making the mask but it was much more time consuming than inpainting. What's worse, you are now restricted to that mask with its exact gradient. So it's a bit inflexible.

 

Typical scenario. Show it to a client and they say. (and I've had this sort of thing happen often) "Fine, but a bit less blur in the middle, I want the marble to show a bit more" you have to recreate your lovely mask again, from scratch.

 

It only took 5 seconds to inpaint the model, and now it is very easy to experiment with different blurs on the background. Change the amount or blur, change the gradient etc. Surely your mask has locked you in, so if you decided you wanted more blur around her feet, or a different amount of blur transition in the middle, you couldn't do it. You would have to remake the whole mask again.

 

On the last one, I could go back and simply redo the filter and gradient in seconds. No need to go to all the effort of making a mask with a silhouette and a gradient again.

 

I agree that a mask might sometimes be best (or necessary) but inpainting is so fast in Affinity. 

 

Flexibility is key !

[R C-R]

1 hour ago, toltec said:

I had a go making the mask but it was much more time consuming than inpainting. What's worse, you are now restricted to that mask with its exact gradient. So it's a bit inflexible.

Don't you have to make a selection anyway when you use the inpainting method? If so, you can just save that as a spare channel so it can be used as a mask any time you need it. So for example, make the mask layer from that, apply it as a selection to the mask layer, invert it, & apply whatever gradient you want.

[toltec]

1 hour ago, R C-R said:

Don't you have to make a selection anyway when you use the inpainting method? If so, you can just save that as a spare channel so it can be used as a mask any time you need it. So for example, make the mask layer from that, apply it as a selection to the mask layer, invert it, & apply whatever gradient you want.

 

Sure, but it's having to make that combined outline shape and graduated mask I don't like. It's "destructive" in a constructive way 

 

As I said, if you need to change the blur later, you have to remake the whole combined mask. If you "inpaint" you skip that whole step.

 

I like simple, fast, flexible techniques. Even if they are unconventional. Too many years in the print trade, you see, everybody always wants it now, or sooner. Can't say I miss it to be honest.

 

However, I do like the flexibility of having several ways to do any job. No method works for every job or suits everybody.

[R C-R]

3 hours ago, toltec said:

As I said, if you need to change the blur later, you have to remake the whole combined mask.

Why would you have to do that? The point of using a mask is you can apply whatever blur you want without it affecting the masked part. Look at the way things are stacked in Jimmy Jack's post. The mask on the "Two" layer prevents the live Gaussian & Lens blur layers applied to it from affecting the woman or the foreground in the background layer.

 

Plus, if you have saved the original selection to a spare channel, if you want to change the gradient part of the mask layer, that is easily done by applying the spare channel as a selection to the mask, inverting it, & adding a new gradient (or whatever) to the mask. The inverted selection prevents the gradient or whatever else you want to do to the rest of the mask from affecting the part masking the woman.

 

I just did this with a more complex photo with two people in the foreground. It took a few minutes to get the selection exactly right to mask them out, but once I did that & saved it as a spare channel, everything afterwards was totally non-destructive & editable because I was using live filters.

[toltec]

OK, I will go through the procedure again and see if I can get my processing faster. It took me quite a bit longer last time and I'm fairly used to making masks.

 

It's always nice to be able to use several techniques for the same job.

[R C-R]

1 hour ago, toltec said:

It's always nice to be able to use several techniques for the same job.

As a very crude example of the flexibility of using a mask for this kind of work, I made the attached using your original image from this post. Please keep in mind the source is a low resolution jpeg & I was not super careful about making the selection tightly conform to the outline of the model, which is the source for the 'photo background' spare channel & the basic mask before the gradient was applied to it.

 

It includes two live blurs, each of which can be edited or used independently of each other, plus a couple of artistically questionable adjustments added to the original Background layer, included only as an example of how the model could be made to stand out even more from the background.

 

At least for me, this method is quicker & easier than trying to do all the inpainting/cloning, hiding, etc. of the alternate one you proposed, & gets more controllable & predictable results.

original.jpg.afphoto

[MEB]

That doesn't solve the issue @toltec is referring - the inflexibility of the combined shape and graduated mask. To get the most out of this you must separate the two masks. Besides that, if you use a Fill layer (gradient) for the graduated mask you can adjust the graduation/fading of the filter non-destructively using the gradient handles/controls. See the attached file for an example: mask_MEB.afphoto

[R C-R]

18 minutes ago, MEB said:

That doesn't solve the issue @toltec is referring - the inflexibility of the combined shape and graduated mask.

True, but that is why I suggested saving the selection to a spare channel before creating the mask or adding the gradient to it. That makes it possible to edit the mask with the spare channel applied as a pixel selection, thereby masking out the model so applying a new gradient (or whatever) will not affect it.

 

It is true this does not allow adjusting the gradation non-destructively, but I just usually find it easier to work with a single mask layer, particularly when I want to do something a bit different than just adding a gradient fade to the mask. I will also sometimes create more spare channels from various selections & use them in combinations to create different pixel selections to include/exclude different things from the mask, or from other layers in a more complex, multilayer document.

 

Like toltec said, it is always good to have several techniques available.  

[JimmyJack]

My point was (and is) that the inpainting isn't necessary. The combo mask was just to illustrate that point in one cohesive image, and yes it's pretty locked in.(In fact it was an extra couple of stesp to combine two masks together.)

But it doesn't have to be that way at all. In practice it's every bit as flexible as you describe.

 

All you really need is a mask of the person, just like you did in your first post.

What I'm suggesting is that instead of using that selection to copy out the person into a new layer in order to inpaint on the full image below...

Just use it to mask her so the blur doesn't effect her (use preserve alpha). Doing this will eliminate the possibility of pieces of her bleeding out into the blurred background.... which is what I was considering to be the primary problem at hand. The blur, of course, can be defined and redefined by a gradient over and over again to your heart's content.

 

But if you enjoy inpainting, and it doesn't take you long, go for it . I'd rather not do it at all. I'm lazy like that .

[MEB]

1 hour ago, R C-R said:

True, but that is why I suggested saving the selection to a spare channel before creating the mask or adding the gradient to it. That makes it possible to edit the mask with the spare channel applied as a pixel selection, thereby masking out the model so applying a new gradient (or whatever) will not affect it.

And this is better than having the separate mask available directly in the Layers panel ready for editing? ok.

[R C-R]

Just now, MEB said:

And this is better than having the separate mask available directly in the Layers panel ready for editing? ok.

As I said, I usually prefer to work with a single mask layer, using selections as needed when I want to edit some part of it (not necessarily just a gradient), or when I want to use a selection most easily made on a single/combined mask layer on some other layer, or for other reasons like to reduce the layer complexity of a document with many layers or simply to make it slightly easier to toggle on & off one mask instead of two when I want to do quick comparisons.

 

It is just one of several techniques I believe are worth considering, like the others with its own pros & cons that will make it better for some purposes & worse for others. So although I may not have been as clear about it as I could have been, all I was really trying to say is it never hurts to consider all the alternatives.

[toltec]

1 hour ago, JimmyJack said:

My point was (and is) that the inpainting isn't necessary. The combo mask was just to illustrate that point in one cohesive image, and yes it's pretty locked in.(In fact it was an extra couple of stesp to combine two masks together.)

But it doesn't have to be that way at all. In practice it's every bit as flexible as you describe.

 

All you really need is a mask of the person, just like you did in your first post.

What I'm suggesting is that instead of using that selection to copy out the person into a new layer in order to inpaint on the full image below...

Just use it to mask her so the blur doesn't effect her (use preserve alpha). Doing this will eliminate the possibility of pieces of her bleeding out into the blurred background.... which is what I was considering to be the primary problem at hand. The blur, of course, can be defined and redefined by a gradient over and over again to your heart's content.

 

But if you enjoy inpainting, and it doesn't take you long, go for it . I'd rather not do it at all. I'm lazy like that .

My point was (and is) that the masking isn't necessary. 

 

I timed it and inpainting took 10 seconds, much of which was Affinity working on it on my elderly i3. I suspect it would be much quicker on my i7.

 

And, I can now use that background to paste in several shots of the same model. Or a different model. No masking required.

 

But if you enjoy masking, and it doesn't take you long, go for it . I'd rather not do it at all. I'm lazy like that .

[JimmyJack]

20 minutes ago, toltec said:

My point was (and is) that the masking isn't necessary. 

 

I timed it and inpainting took 10 seconds, much of which was Affinity working on it on my elderly i3. I suspect it would be much quicker on my i7.

 

And, I can now use that background to paste in several shots of the same model. Or a different model. No masking required.

 

But if you enjoy masking, and it doesn't take you long, go for it . I'd rather not do it at all. I'm lazy like that .

 

Boy, I don't get it. How did you separate the woman then? 

That selection is your mask, there is no extra work.

Guess I've saved myself 10 seconds .

[toltec]

57 minutes ago, JimmyJack said:

 

Boy, I don't get it. How did you separate the woman then? 

That selection is your mask, there is no extra work.

Guess I've saved myself 10 seconds .

 

I've tried but every time I apply a lens blur I get a halo. 

 

The mask is easy, I can create that from Refine edges but it does not mask the halo. Which adds a nasty edge around the legs, especially.

 

Did you use a Live Filter Layer, and if so how did you stop the halo. I can get OK results if I apply a lens blur, but not a live filter blur. Which was the original issue, hence the inpainting.

 

Maybe with all the 10 seconds you have saved, you could explain the steps you used. Please !

 

 

 

[JimmyJack]

27 minutes ago, toltec said:

 

I've tried but every time I apply a lens blur I get a halo. 

 

The mask is easy, I can create that from Refine edges but it does not mask the halo. Which adds a nasty edge around the legs, especially.

 

Did you use a Live Filter Layer, and if so how did you stop the halo. I can get OK results if I apply a lens blur, but not a live filter blur. Which was the original issue, hence the inpainting.

 

Maybe with all the 10 seconds you have saved, you could explain the steps you used. Please !

 

 

 

 

Sure. Will do. But MEBs file is a fine example. The only thing I (might) do a little different, is to draw the gradient on the blur filter itself rather than using the fill layer (not that there's one thing wrong with that at all).

 

While I'm putting together some screen shots, can you post an image of the halo?

 

 

[toltec]

Ta

[R C-R]

I don't get it either. Among other things, if I remove the model from the background layer by using inpainting or cloning tools to fill in the large area covered by a foreground subject like a model with anything sampled from the rest of the photo, unless I spend a lot more than 10 seconds tweaking the results to get something that looks reasonably natural & does not include obvious duplications from the rest of the photo, I do not see how that background could be used with other shots.

 

At least for me, those two tools generally don't work very well on areas that large unless I spend considerable time building them up by applying them to small overlapping areas with soft edged brushes & other techniques to eliminate repetitive patterns & other obvious artifacts. Even then, the result often looks too artificial to use for much of anything. Maybe I just don't have the skills for that but I suspect I am not the only one who would need considerably more than 10 seconds to get anything that could be used with other shots.

[toltec]

2 minutes ago, R C-R said:

I don't get it either. Among other things, if I remove the model from the background layer by using inpainting or cloning tools to fill in the large area covered by a foreground subject like a model with anything sampled from the rest of the photo, unless I spend a lot more than 10 seconds tweaking the results to get something that looks reasonably natural & does not include obvious duplications from the rest of the photo, I do not see how that background could be used with other shots.

 

At least for me, those two tools generally don't work very well on areas that large unless I spend considerable time building them up by applying them to small overlapping areas with soft edged brushes & other techniques to eliminate repetitive patterns & other obvious artifacts. Even then, the result often looks too artificial to use for much of anything. Maybe I just don't have the skills for that but I suspect I am not the only one who would need considerably more than 10 seconds to get anything that could be used with other shots.

 

Well, maybe not that shot because of the column and the paving but I've done it often enough with other stuff. Especially if blurring the background.

 

My real issue is the halo whenever I use a live filter layer. 

[toltec]

By maybe I mean definitely 

[JimmyJack]

Okay, so.... I can't see how to add images inline with the text, but at least they're in order.

 

1) Same as you. Use selection brush to make initial selection and refine.

 

1b) Using the Mask output. This is what I get.

 

1c) .....with the background showing through.

 

2) add a/any blur filter

 

3) draw gradient right on blur. Reposition-able.

 

 

[MEB]

 

@JimmyJack The only reason i have not applied the gradient to the filter's buit-in mask is because it's not editable. If you need to adjust it after deselecting the filter layer, you have to rebuild/adjust the gradient from scratch again.

[JimmyJack]

8 minutes ago, MEB said:

 

@JimmyJack The only reason i have not applied the gradient to the filter's buit-in mask is because it's not editable. If you need to adjust it after deselecting the filter you have to rebuild/adjust the gradient again.

Ah yes, I forgot it has no memory (oh the irony). Redo-able yes, but no memory.

A box (or any shape) retains the gradient controls too...

[toltec]

JimmyJack

 

Thank you.

 

I shall try that tomorrow as soon as I have a proper pc to hand.

 

Obviously you didn't get the halo that way.

 

Unless its the lens blur that's the issue? I notice you used Gaussian