AI generative Fill in Affinity

[CM0]

6 minutes ago, R C-R said:

So about the only thing newish that is revealed about it is that in its 'pre-alpha' state it can run Javascript code. 

I mean that is an extremely minimized view of the significance of that post. It is huge reveal that there is a concrete implementation in progress using a modern JS engine. That communication and demos were sufficient and very well received if you follow the rest of that thread. There is no need for a timescale for its release. Showing progress is far more effective than anything else they could ever post and the solicitation of feedback in that thread was very welcomed.

[evtonic3]

Can someone please post the links mentioned? Sorry if you already shared. In English?

[Granddaddy]

1 hour ago, Frozen Death Knight said:

Got a link to the article? Interesting!

It's a post a few above mine but it's getting lost in the chatter. The PDF of the translation is at

https://forum.affinity.serif.com/index.php?/topic/186676-ai-generative-fill-in-affinity/&do=findComment&comment=1144217

 

[Affinityconfusesme]

 I would guess that these things would come in 2.5. The demos were from the end of march. So presumably they are a lot closer to releasing it for beta than you think. 2.5 would make sense although not guaranteed because that would be in around 5-6 months from now if every if 2.4 is 2 months also like 2.3 and so on.

[R C-R]

58 minutes ago, CM0 said:

Showing progress is far more effective than anything else they could ever post and the solicitation of feedback in that thread was very welcomed.

Effective at doing what specifically? All they have said is they are working on what is still an early pre-alpha version & had to reiterate in several replies that they cannot even promise that any of the things demonstrated or requested will make it into a beta or retail version.

[Frozen Death Knight]

31 minutes ago, Granddaddy said:

It's a post a few above mine but it's getting lost in the chatter. The PDF of the translation is at

https://forum.affinity.serif.com/index.php?/topic/186676-ai-generative-fill-in-affinity/&do=findComment&comment=1144217

 

So to summarise:

- Because of the growth of the Japanese market for Affinity the devs are now working on "vertical writing" and "ruby" to support Eastern languages. Good news for a lot of people I think.

- A cloud service to be able to easily switch between platforms and keep all your work assets. Will help with a steady revenue stream and I personally think is a reasonable way of monetising the Affinity platform without backing away from comitting to perpetual licenses.

- AI functionality has been worked on for 2 years, but nothing concrete about what that feature is. Kind of implying that it will be released next year, but that's just my speculation.

- Any of these are highly likely to be officially announced within the next 3-6 months.

So we have some idea of what to expect for 2024. Sounds good to me!

[CM0]

21 minutes ago, R C-R said:

Effective at doing what specifically? All they have said is they are working on what is still an early pre-alpha version & had to reiterate in several replies that they cannot even promise that any of the things demonstrated or requested will make it into a beta or retail version.

Suffice it to say many found value as demonstrated by the responses in the thread. I don't think I can explain any better than I already have. 

[Jörn Reppenhagen]

I am badly missing two specific topics in this AI discussion: 1) AI, 2) Cloud or local AI use?

As said before, PS and LM voted for a cloud approach.
Benefits are quick and easy implementation without specific hardware requirements, no need to download and install huge data volumes prior to use.
But the drawbacks are serious:

• NO PRIVACY anymore. AI needs to know the whole image to decide which AI model to use - there are countless AI models for multiple purposes, each providing it's own picture style; thus the model needs to be chosen or adapted according to the image to process. In addition, each and every image needs to be scanned for prohibited contents, so the cloud service can deny processing your image. Thus each and every image we process using AI needs to be transferred to unknown destinations.
• Massive restrictions regarding contents, especially sensitive if it comes to any sexual-related matters. I saw AI models denying to generate anything containing terms like "butt". So you need to become real creative if you wish to remove anything near such body parts. Also many AI models restrict celebrity rendering or any styles similar to styles of famous modern artists, also copies of other works. Of course there's a high risk of false positives.
• Restricted flexibility: There's a wealth of great extensions for AI generators like Stable Diffusion, free of cost. You cannot install such extensions in a cloud environment.
• Endless operating costs. A cloud solution would force Serif to introduce one of the dreaded subscription models, the reason why we abandoned Adobe, the reason why we chose Affinity. Of course AI features could be offered as a separate service - but this would lead to the same ugly result: Subscription. In that case, we could also return to Adobe. And that's exactly what would happen.
• Legal liability if abusive images get rendered by the provider's cloud service. Thus the provider needs to examine every single image very closely, worsening the privacy issue mentioned above, also increasing risks for the provider.

That's why I believe a local install would be the only way to go.
The question is not if Serif needs to implement AI features, the question is how this should be done in the best way possible.
I guess it would be quite beneficial if we try to collect some ideas helping Serif with their decision.

[R C-R]

5 hours ago, Jörn Reppenhagen said:

The question is not if Serif needs to implement AI features, the question is how this should be done in the best way possible.

There is also the question of which AI features it should implement.

5 hours ago, Jörn Reppenhagen said:

I guess it would be quite beneficial if we try to collect some ideas helping Serif with their decision.

Maybe they would consider those ideas but it is very unlikely they would even notice them here in the Questions forum. Those suggestions would be much more likely to be considered if they were posted to the appropriate Feedback & Suggestions forum.

[Jörn Reppenhagen]

17 hours ago, R C-R said:

There is also the question of which AI features it should implement.

That's a secondary thought as the heart of most AI features is the models/checkpoints and a local architecture including a Python installation.
If you wish to use generative infill, inpaint, outpaint, upscale, importing custom objects/faces into the database, even swap things (or faces) in movies and the like - it all builds on the same base infrastructure. And it all works with a local installation, without sending all your personal works to foreign servers, keeping your property your property.

Quote

Maybe they would consider those ideas but it is very unlikely they would even notice them here in the Questions forum. Those suggestions would be much more likely to be considered if they were posted to the appropriate Feedback & Suggestions forum.

That's also a secondary step as a suggestion should be something we are sure of. I don't see too much sense in drowning a suggestion in countless pro and contra posts, Serif guys would very quickly lose any interest in following an endless discussion. Thus I find it better to discuss a suggestion and come to a mutual conclusion, THEN launch the idea as a suggestion.

[R C-R]

5 hours ago, Jörn Reppenhagen said:

That's also a secondary step as a suggestion should be something we are sure of.

There is no way we can be sure a particular suggestion fits with their plans or what they are already working on, to say nothing of the near impossibility of getting a consensus among such a diverse user base as this about what should be implemented or how.

Please at least consider posting suggestions to the forum purposed for that. There is just about zero chance that anything mentioned here in the 18 page topic will be seen by the developers.

[Jörn Reppenhagen]

20 hours ago, R C-R said:

There is no way we can be sure a particular suggestion fits with their plans or what they are already working on, to say nothing of the near impossibility of getting a consensus among such a diverse user base as this about what should be implemented or how.

Of course there's no universal way - but that's the exact purpose of discussing such a topic: Finding out what Affinity users actually want - and in which way this is congruent with Serif's plans. But that requires talking about possible implementation options, their drawbacks and benefits.

Quote

Please at least consider posting suggestions to the forum purposed for that. There is just about zero chance that anything mentioned here in the 18 page topic will be seen by the developers.

I already answered that in my previous post. 😉

[thomaso]

1 hour ago, Jörn Reppenhagen said:

that's the exact purpose of discussing such a topic: Finding out what Affinity users actually want

Consider the existing discussion … and your initial handling with existing views:

On 11/14/2023 at 1:05 PM, Jörn Reppenhagen said:

I didn't read the full discussion (16 pages is a bit much), 

… followed by simple + general prophecies about others and about all users as a homogeneous mass:

On 11/14/2023 at 1:05 PM, Jörn Reppenhagen said:

I didn't read the full discussion (16 pages is a bit much), but here's three points to consider:

1) Looking away (or making fun of the technology) does not help at all. People WANT it.

2) Everybody (…) will LOVE this new tool.

3) Serif will lose (…)

Apart from this odd way of communication:
Serif is able to consider various aspects either without vague or precise opinions (which aren't unique or any special in this forum) but they also can …

• actively ask forum users for views on future developments (as done for "Scripting" and "Global Layers" for instance) and

• contact users via email (not the minority of forum members only) with a link to an online survey/poll (as done in July 2022) and

• launch an extra beta program to discuss with users (as done in Feb 2022 for V2).

[KCP]

Someone on this thread mentioned the potential end of Affinity if it doesn't adopt generative AI. As a commercial user of Adobe for several decades and a recent convert to the Affinity suite, I find no reason to worry about that. I was a power user of Adobe in the brand identity field and eventually decided to switch after retiring. With no more need to regularly share files, I was more than happy to cancel my overpriced membership. In a nutshell, Before converting, I ran three months of extensive comparisons to ensure that Affinity had the requisite tools and sophistication to compete with Adobe. The short answer is that it more than does that and a bit more. Adobe's AI has some value on a mass production level, although it's unnecessary for essentially all of the commercial workflow I ever encountered in 40 years of commercial work. 

Many of Adobe's newer tools have been strategically designed to automate processes for the average consumer and mostly bloat the apps with unnecessary automation. As someone with many years in and around advertising, my suggestion is not to get sucked in by shiny objects. Affinity is a powerhouse and getting better with every update.  

[smallman]

I will buy Photoshop in the future. if Affinity photo dosen't support AI Generator. end of story.

[_考槃]

我认为这种Ai功能应该是对传统工具的解放。

[nezumi]

Great example of AI superiority when it comes to cropping. One click and done.

If you dont needed that hat and leg that is... 🙃

[spacemishka]

1. Well it deserve a subscribtion modell to pay for the AI service

2. get your photo right and you don't need to worry much about post processing...

[ShinobiRetro]

On 5/24/2023 at 4:16 AM, WMax70 said:

Wondering how Affinity Photo is preparing or has prepared the integration of AI like Adobe does at the moment. 

I saw that Adobe Photoshop is now able to work with AI. Insert objects with written statements, remove items enhance the image and edit in seconds in stead of hours. Ai is now entering the photo tools.
It’s amazing and I’m looking forward to get such also in Affinty Photo.
It is for me worth to switch back since my workflow and edit time will be substantial reduced. 
Have a look at it below and just like the title says, it’s amazing. 
Serif, Can we expect such also in short term? Is is already available as plug-in in a certain way? or will this be End Of Life of Affinty Photo? 
 

 

This will most likely be a key feature of Version 3.

[albertkinng]

On 12/4/2023 at 10:29 PM, smallman said:

if Affinity photo dosen't support AI Generator. end of story

We will miss you.

[Juan Garcia]

On 5/25/2023 at 5:01 PM, dominik said:

Thank you @carl123 for posting this image.

In contrary to all the praise about AI I want to point to a single screenshot of this video (where the presenter says "What, that's insane!"):

 

Yes, Photoshops new 'generative fill' (and on step before 'select subject') did do a decent job in replacing the background in this image. But at the same time it is rather easy to tell that this is a montage because the lighting around the model's arms just doesn't fit well with the overall lighting of the background. I would expect a montage done by artificial INTELLIGENCE a lot more smarter. This is not an insane result!

To me this is a typical excample of all the hype without looking closer on the real results.

I concur with the assumption that this technology will refine and mature over time and as mentioned in this thread it can become a true help to tedious tasks like image composition.

But at the same time, please, don't get overexited about another video that is nothing more than an Adobe advertisement 😉

d.

For me the real usefulness is if you want to take her hair and move it behind her shoulder once the model is no longer there. And smaller things that most editors like Pixelmator Pro and Luminar Neo can do, like to remove her tattoos. 

[rfjphoto]

After watching Capture One videos on AI I have decided against Affinity, I was considering the software, but not having AI is my reason not to. I am justy one person, but I am sure more will make this decision when the see what AI can do, and how much time it saves in doing masking alone.

[spacemishka]

an API connection to Stable Diffusion would be ok for me.....

[Juan Garcia]

I bought PixelMator Pro, and I am finding myself not using Affinity Photo at all. I won't be buying the next version and to be honest I doubt there will be a next version anytime soon. There's not even regular updates anymore...

[GarryP]

12 minutes ago, Juan Garcia said:

I doubt there will be a next version anytime soon. There's not even regular updates anymore...

What is it about this release schedule which suggests to you that there’s “not even regular updates anymore”?

• 2.1.0 – May 18, 2023 – and then about one month until,
• 2.1.1 – June 15, 2023 – and then about three months until,
• 2.2.0 – September 14, 2023 – and then about one month until,
• 2.2.1 – October 17, 2023 – and then about one month until,
• 2.3.0 – November 29, 2023 – and then about two months until,
• 2.3.1 – January 8, 2024 – and then about one month until,
• 2.4.0 – February 26, 2024 – less than two weeks ago,
• 2.4.1 – Already being worked on