drstreit

Nah - I am just lazy and I only dabble in the arts as a hobby and thus need easy tools But my professional life very much has focus on AI, I lead building global projects using AI models for the life science industry - and I see a frightening change of the people coming from university: They simply DO NOT ACCEPT that there is a long, manual way that gives you 100% of the desired result in many hours of time. Instead there should be an app like way to get your pareto 80% with two clicks: Good enough is the way to go agile. And with that approach, this huge oecosystem of grafic design tools were created on the various mobile devices - with functionality that is astonishing good nowadays with only seconds of effort. The easiness how these kids ("kids", in my eyes, I am over 50 now) accept wonders of software such as AI models deliver today - only to request more of it and and faster - led to the disrupting push of AI into every industry. Mark my words: Either all the software giants (and mini-giants like Serif) adapt - or they will die out with only us dinosaurs doing things "the right" way. I myself am very impressed creative minds such as Kevin Hess who shows us that an author with self proclaimed small grafical capabilities can produce a full grafical novel by his own with the new AI tools: https://beincrypto.com/ai-art-worlds-first-bot-generated-graphic-novel-hits-the-market/ And sure one can complain about the style and limitations - but Midjourney (the AI he used) has been released mid of July this year: Just imagine what these tools will do in 2, 5 - 10 years from now given that AI complexity has been exponentiell the last decade. I would dream of Serif trying to spearhead that development with a maximum aggressive move: In my professional opinion (admittingly one with only experience from other industries), Serif does not have the time to wait another year to move forward.

No, that is plain wrong: AI models work a bit like our brain, building billions of weighted connections based on learning. With this, they are capable to deduct new solutions to problems that they initially were not trained on. As they are missing context, its the task of the data science engineers to steer them, the constantly improving models are the result. Say I train a model on 90 different species of dogs, it will be able to identify all 300 or so of them, deducting the characteristics. It gets better: Reinforced learning models constantly create new ways to solve a problem or draw a picture, rate the result and learn from it. The alpha models from OpenAI beat every chess and every Go player on the planet: But they never actually were fed any strategy books, they derived their strategy themselves from billion of matches against themselves… Better believe that well trained AI models are creative in the everyday meaning of that, we are there to give them context.

No, I dont validate anything with the simple fact that tasks will be replaced by AI. That is a nobrainer, or do you still use paper maps driving around - or do you rely on a navigation system? Things will simply change: With AI being integrated great tools - such as Serif's - you can work fast, produce more, automate simple tasks. Just imagine to train your unique style to an AI and produce your own grafical story within days. The whole point of this forum post is that I believe AI will greatly help designers - and I dont want Serif to miss that train, because simply people will leave.

Will do - sorry for my enthusiasm. Coming back to topic: I experiment with https://barium.ai/dashboard at the moment as I need seamless patterns for some 3D work - would love to hear your thoughts. And exactly this application would be a great first addition to Serifs toolbox, as models for patterns might be less controversial in terms of copyright and actually needed quite often!

Tricky: Currently the trend is more understanding your personal DNA to design medicine that is tailored to your problem - e.g. what we call "cancer" is actually extremely specific to the host. And the advantages are immense - have a look at Alpha Fold - I am not to dramatic to call this the holy grain of medicine: Understanding how proteins are folded is the key for developing treatments for all things involved. There is a reason I am a firm believer of AI - but I am also a believer for strict control, as AI usage can be dangerous, and will continue to grow more dangerous with the growing power of the models - if you want to read a actually quite realistic SkyNet scenario, just read https://www.washingtonpost.com/opinions/2022/08/31/artificial-intelligence-worst-case-scenario-extinction/ ... Having said that, generating images with AI is not too dangerous - but fun with no end: Running an anime contest with my daughter at the moment who can create the best short story with the generated pictures - hmm, actually I am loosing ...

Sorry - too late with the cooking: https://researcher.watson.ibm.com/researcher/view_group.php?id=5077. But in earnest: Printing books destroyed the scriber class of the midle age, photography destroyed most of the painter jobs that normally were called once a year into the homes to paint your family, around the corner where I live the last photograper of my small town just closed down: I guess smartphones and easily available full formats killed his business... Yes, a lot of designers jobs WILL be destroyed through the AI revolution we see since a few years. As will such diverse jobs as technical writers, lawyers, etc.: At the end EVERY job that has a huge amount of routine work WILL be replaced by AI - no question around it. What will stay are jobs that are just not worth to automate because that work is too cheap jobs that do have a high amount of creativity and personal involvement After each of the revolutions mentioned above, the number of total jobs actually grew, even exploded: It will be the same with this one. If at one point we actually have not enough meaningful things to do for all humans in the world, the things might get tough: But honestly, I dont see that situation in my or my kids lifetime.

Absolutly, that is already a problem: Just think about the Deep Fake images/videos that are available, plus of course - humans being what they are - Deep Nude Fakes... A couple of countries outlaw the creation of certain pictures, but that is quite difficult to define at the end: Should a caricature be banned in principle, because it shows a living person? Its about the usage scenario and the intention I guess. At the end, as with every new technology, there is a need for a broad discussion - at the moment the main drivers are AI specialists and technicians...

Disclaimer: I am complete amateur with creating seamless patterns, so a question to your expertise: Would you mind to have a look at https://barium.ai/dashboard , maybe register and try out some patterns and tell me if that is something you see as being useful? It uses the Stable Diffusion model (the only one to my knowledge that you can easily deploy locally), so its quite accessible and extendable.

Good reference! Yes, I also see the copyright problem being unsolved - and another huge challenge is also the fact that you can reverse engineer these models, to get back the training set data: Might not be too critical with art, but I work with health data and AI models... Of course, I try to be the AI's advocate here in this post, so a bright side of these models: One can suffer the fact that these models do train on existing data - maybe even yours. Or you can make the power of the models your own asset: It needs surprisingly few pictures of a given style to train the mdoels to recreate that style. That means you could easily extend the models with your own style (even locally, not making that public), and then utilize a model that mimics your own style for your work - adding unbelievable speed to your production pipeline...

Here the link to the training set of pictures currently used - you can check there if anything from your portfolio was used - and you can exclude it: https://laion-aesthetic.datasette.io/laion-aesthetic-6pls/images?_sort=rowid

Very much uncharted territory here: Would you sue an artist that looks at your pictures, gets a creative vision and goes home and starts painting? Can you really convince us, that you NEVER looked in your life on the work of others and somehow developed your unique style with that impressions as a basis? And its also the other way around: AI generated picture also can have their IP protected - see here! One big failure of any IP discussion is the acknowledgement that we all only develop EVERYTHING in our life while copying others behaviour, ideas, art, whatever. That does not make it right to produce a 1:1 copy of anything - but it will grant you a new IP if the innovation/creativity barrier has been crossed: And that definitely is the case in AI generated images if you compare them to the training sets...

Great text - and I fully agree! Hence this post actually - I would really hate to see Serif fall because an industry trend was slept through (not only talking about AI, SDKs in general are must-have nowadays to be successful). Unfortunately, this seems to be the case now: Without plugin interface, we cannot be active as community - with one, the number of plugins and tools would explode. Its not even on the roadmap, so I fear its at minimum a year ahead of us: I just hope that is not too late.

I agree, a full fledged API would snowball all kind of plugins, AI included. Even Adobe cannot create all content by themselves, so the rely on the community. I teceived the same feedback from Serif as you mention: „Long term idea to have a SDK, no roadmap yet“. I struggle to understand how one can release ANY software product nowadays without automation API… Seems like shooting your own foot a bit…

Great example here, DeIN! I honestly struggle how anyone fails to see the upcoming impact: These models are only released some weeks ago, looking at the speed they evolve, we are seeing a freshly hatched 🐣 here that will turn into a man-eating ostrich within years… Lucky the one to have tamed it then… And everybody: Give them a try, they are honestly just great fun to experiment around!

Come on guys, you remind me of the stories of the first steam train passengers, who fainted from the neckbreaking speed of 20 km/h and were convinced that humanity would never strive to move that fast… So the pigs dont drink? Just take them as PhotoShop 1.0, Microsoft Windows 3.1 or a Assembler Compiler against C#… These tools are incredible powerful today, weeks after their release. But they are already good enough to generate new images to again train them to get better, allowing Reinforcement Learning with billions of new connections per day. And they show only percentages of what already exists, as the full models are not public, with especially Google not releasing because its „ too powerful“…

Because these things are only weeks old and are only starting to develop. New features will be outdrawing existing pictures while copying style and extending content, adding content aware parts into images etc. A very close workflow will define how efficient that is: With your argument, every use of filters would be useless, as you can always use a standalone software. But we use integrated filters as that workflow is more effective. Back to the topic opening: Serif does not have to do anything, beside giving us an SDK. Not doing it will hurt the tool suite the long run. Nobody forces anybody to embrace or not AI in imaging - I just brought my daughter to her horse riding lesson, so enough people still seem to preferring horses to cars…

You refer to a very specific part of working with images - I would argue that a professional Speedtree modeller does indeed see him/herself as an artist and the outcome does sell... So again: I don`t want to argue art here - I acknowldege that many visuabl products do sell across the industry - and I just ask for a VISION to allow Serif users stay on top of the curve here through integrating new tools into the Affinity suite. Can be through a vendor integration , can be through releaseng a SDK. Preferrably the later - are there really vendors left that do not understand the power of user created content? Would have Photoshop ever reached its power without constantly people integrating it into their workflows or creating filters through their (pain to work with) SDK?!

Nothing wrong about it - still something you might want to use Affinity products for - thats what the post is about, enabling excatly that!

In my humble opinion, you comfuse the current "AI-on-every-tool trend" with the current release of general purpose models. I am totally with you on the first - "AI" ob everything just relableing old stupid routines mostly. Here I see some progress especially on mobile platforms with face parts recognition, but really nothing groundbreaking - its more marketing than anything else. A total different story are models such as DALL-E, Stable Diffusion, Midjourney etc: These models are trainied with billion of pictures and are able to put out new things - new in the sense of creative (of course just recombining - but that is the reason I asked for having them as tools for us to guide them within the Affinity suite). Your same arguments were done with rather similar AI models playing chess, playing go, translating or creating texts etc. - with known results... Below some pictures that I just tried to put together for this post (and I am actually quite bad at this currently): I fully trust that you could easily do far better on each style - but can you do it within minutes with hundreds of variations? Its not about stupidly generating pictures from scratch (ok, I admit - its actually really fun doing so 🙂), its about asking the model to create parts of it, combining it, merging it - at the end nothing different than a specialized brush - only more "intelligent" And last sentence: Looking at what art is currently soldfe, I habour some doubts if your expectation of creativeness, exploring totally new fields of imagination, boldy go where noone has gone before that you demand from our field is really the barrier that every artist has to face these days...

Here a nice example what people start doing with AI generators (beside cretaing more cude cats 🐈): https://beincrypto.com/ai-art-worlds-first-bot-generated-graphic-novel-hits-the-market/ And it also gives a vision how a tool suite like Affinity could deeply integrate it into the workflow for production, optimization, publishing.

Hmm - the name of the forum is "Feedback for Affinity Photo on Desktop " - so let us hope they at least read it. At the end, we are trying to help - and just plainly ignoring customers is rarely the way to success from my experience...

I missed that one: Care to share the URL?

Let us go back to the topic: *Affinity: Do you plan to support the various AI generators within your product suite? Maybe even spearheading that movement instead of forcing us to go to different products to try it out?

I feel you are not too enthusiastic about AI here - but honestly: Every new technology has every changed some job profiles (or even vanish them). Sure, in 5 years you will be able to tell an AI: "Draw me a 100 pages comic around the story of xyz, don't forget that I love cool sword fights and a happy end" - and it will be delivered in minutes... Is that good or bad? I tend to see it as good, as every human progress in the end is a good thing from my point of view - but it will take years to adapt: society, job profiles, (mis-)usage etc. What is the saying? "Society tends to overestimate the short implications of innovation, while underestimates the long term effects..."

Great post, PaulMc! I wholehearty agree - I also do not see AI as replacement, but as an important part of a pipeline of the future (even soon for vector grafics: That is due to the early stage of the software, but seperating the results into vectors should actually be quite easy, shouldnt it?). Still the question is open: What are Serif's plans for the future here? And sidenote: Not really answering the communitys question about SDK access for months (beside: "Not planned") is not really something I do understand...