@ExplorewithZac
This is a really interesting talk to me as an genAI developer... I like the change in mindset from "We're trying to automate absolute everything" to "It's okay and GOOD to expect users to physically create the inputs for these models".
@FusionDeveloper
August 22, 2022: Stable-Diffusion 1.4 release date. It hasn't even been out for 3 years yet.
@cybermad64
The real question now is : "How can we emulate the human touch and automate the iteration process"
@CoolSmek
Using ai as a collaborator in generating artwork has been a really fun creative activity for me personally. Generating images based on a prompt with no other control is pretty boring.
@GnaReffotsirk
This is what I was trying to say since that poster making contest made gen ai public
@RachealStahlman
Great talk and perspective, Thank you for focusing your talk on this. I also am an optimist and while I totally understand the concerns people have and their worries are completely valid, I also know that technology is always evolving and staying curious and adaptable is key to maintaining a competitive advantage. My hope is we continue to bring our unique perspectives and talents into the creative process with AI, so we can create our best work but also retain the qualities that make it personal and unique to us and we never replace the human aspect of the creation process.
@silentbrahmin
When someone trains an existing model like SD on their own inputs, that’s actually finetuning the model, right? Which means the base model will still have all the ethical problems, because it’s been trained on scraped data. Are there any models one can use to finetune that’s only trained on data with consent, so that there is close to no unethical activities involved in the model?
@GForce_ART
Well presented. Measured. But, still kind of a sales pitch for Invoke, ultimately. Let's not bs here: The insane rush and push for genAi is to make stock evaluations go up. It put a bad taste in our mouths. One thing that is glanced over here is the simple fact that using genAi may not be RIGHT as a workflow for a certain team culture, product, result, "feel", audience etc. It invariable skips steps in the iteration process where hands and mind have time to mold it. It's basically commissioning out the work YOU yourself would do. The value DOES shift because of this process. It is skipping huge steps that can take away decision making. It's not a tool that helps you along the way. It makes sweeping judgements (based on other people's data, directly). You might get there/get close, but how much time are we gonna have to spend trying to convince a robot teammate to make micro adjustments when you could just make the damn thing... another point is... what if working with genAI is a culture/joy killer and THAT changes the product/work? Are we just genAI tamers now? The game you are making is really just a product of the people working on it. SO if all those people are uninvolved (or mentally checked out and over reliant) and if there is now a "black box" involved, it's like having teammates you rely on that are not actually there. There not even in a conversation. I understand framing it it as a tool...but we ALL KNOW what's gonna happen. Companies are gonna do this the WRONG WAY (already have) and force it, even when it's not the right tool for the job. Slop is inevitable on a scale we have never ever seem. Slop is ALREADY an issue outside the context of genAI, like on Steam. Shovelware, clones, etc. The way the tech was trained is wrong. It just is. It makes it unpleasant to consider using the "tool" in any way. As and artist I love tools. Always open to new tools. But, when we were optimistic about AI assistance we meant spreadsheets and mundane shit, not the actual work, which will be invariably affected somehow as we are less involved. It's good that he is addressing that it's not a "magic bullet", but to the investment parties....it very much is, and there is too much riding on it for them to NOT force it in crazy ways. Their cost is already sunken. This has happened before (like decimating the entire treasure trove of Disney animators after Toy Story came out... and look where they are today...they MAY AS WELL be using genAI for the slop they make. Literally not a brain among them.). Things overall are starting to get under-designed and under-considered. I don't currently see how the proportion of slop to quality is reconcilable. IF I was given the choice between making things sort of faster or have the whole Earth, every ad, every image, the entire internet etc. turn to unmanageable slop heap...not really sure that is a worthy trade off. The reason it feels wrong to accept genAI as a mandatory thing is not the same thing as the loss of the horse and buggy or advent of cotton jin (which is a terrible analogy I hear all the time in comments...). It doesn't make our work easier ONLY. It makes it NOT TOTALLY OURS. And, that is weird and feels wrong (probably because our gut is telling us something). This, for better or worse, will be dictated and mandated by big tech and forced down our gullets with NO regard for the cost. We may be lamenting what we lose... it is totally false to say all change is good. It often isn't. Everything good in this world usually comes from the bottom (before being usurped). GenAI kinda comes from the top, and that's why it smells fishy and has been handled completely wrong. "Efficiency" is not really always the top priority when trying to make something of value. Art has to be slightly wasteful to kind of get to the goods. Having instant iteration is not actually good... because you are not thinking clearly and as engaged. Speed isn't the highest virtue. Never was. Focusing on outcomes at the expense of process will destroy the work, even. Nothing good is made in a rush. Every artist learns this the hard way at some point. It's a delicate balance. SO...good talk, but insufficiently convincing. I think the social and culture cost is already very very clear. It is a high cost, imo. I mean, art thieves are eating very very good right now and the biggest advocates are tech bros, which says volumes to me.
@captanblue
This was great. Right now, my biggest hangup is training this on my own art. It looks like you guys are still working out the kinks.
@swapticsounds
I am doing a slow and iterative process when it comes to making AI music with udio. Often I work many hours on one track and only about 5% of the material I generate within that process ends up in the final song.
@larscwallin
I think the "intellectual" part of IP is kind fundamental 😉
@lodhiayush6718
Someone who spends weeks creating that perfect prompt can not copyright the output that those models generate, Why? Well he could have spent years learning the skills that will empower him to be able to create that work himself Instead he uses models unethically trained on peoples work that they created with years of hard work that got into achieving those skills and you think you are entitled to copyright of a work just because you typed a prompt. Me personally one should not be able to copyright anything even if it has 10% of AI stuff in it, cause when you use even 5 seconds of a song in your video that does not belong to you you can get copyright strike, so the same should stand with AI as well.
@UNIX_OR_DIE
AI is dead! Use our AI instead. <- OMEGALUL!
@paulhiggins5165
Re the definition of 'tools' I think it's under appreciated that even creating Art with a tablet is still a craft skill. There's a tendancy to think that because the artist is working with a computer they are no longer engaging directly with the work. In reality this is not the case- the digital painter or digital sculptor is in fact directly manipulating the work via physical gestures that are used to shape the final result. And note that this is a non verbal process. A major limitation of AI Art is it's dependance on text based interfaces where the intent of the artist must first be defined in words before it can be interpreted by the AI. But while it might be true that 'a picture paints a thousand words' a thousand words may sometimes not be enough to precisely define a single picture. I get that your software attempts to solve this problem by incorporating the use of images as prompts which does mitigate the problem to some degree -but in the end the use of AI will always involve a trade off between getting what you want and taking what you can get. AI will always be a compromise as far as control goes, and this is not a bug but a feature- after all the entire point of using AI is that you don't have complete control- if you did the AI would be adding zero value to the process. So anyone waiting for tool over which they have complete artistic control will be pleased to know that such a tool already exists- it's called a drawing tablet.
@amig012
interesting tool, but free trial requires payment info, instant turn off
@paulhiggins5165
The term 'AI Slop' is not a quality judgement but an issue of provenance- it's not about how 'good' the AI Image is, it's about the fact that the image was created using AI. This is the real challenge that AI generated content will face in the future. AI generated content is paradoxical in that it's virtues- it's speed and relative ease of creation-are also it's vices in the minds of many people. The problem I see with AI is that while it can create 'content', it cannot create 'value'. This is a slippery distinction but it's an important one because value is not an intrinsic property of the content itself, as many people assume. Value is basicly about what people think. An AI generated image might be genuinely pleasing to look at, might be beautiful even, yet still have no value because it was created by AI. If we- as a society- take the view that AI generated content is worthless slop then it will have no value in our society no matter how aesthetically pleasing it may be. This has important commercial implications because if I am using AI generated content in an effort to add value to my products, but my customers take the view that AI content is worthless, then not only am I NOT adding value to my product I may even be damaging that product in the minds of my customers. It might be that the kind of 'hybrid' approach you are advocating might overcome this problem of 'low effort=low value' but I fear that your efforts are being undermined by the AI developers themselves who market their tools as being low effort and low skilled tools. The fact that AI Art is seen as parasitic on human creators adds to this general perception that AI generated art is low value slop and may turn out to have been an error on the part of the developers. AI in general is a PR disaster in terms of how people view the technology and the term 'AI Slop' perfectly represents this negative view. As a content creator I would be forced to think long and hard about using AI Generated content not because it technically bad but because the negative perception of all things AI means that anything created with the help of AI might be seen as being so low value that no one would want to purchase that content.
@BoyanOrion
Thank you for the amazing talk. Everything was right on point and very inspiring! I would like to know if you have any plans to integrate video models like Wan 2.1 in near future? It could be a good addition to the software especially for doing image 2 video or character consistency image sets via character rotation lora. I'm sure you are aware that some amazing tools have just been released recently for this model like specific control nets, inpaint capabilities and other good things. It would be an amazing workflow if you can do image generation and then use that image to create a video, all done in one place, in Invoke.
@lfpfxf
Please give us support for rtx 5080 , cuda 12.8 is not being recognized🙇♂🙇♂🙇♂
@brockoala2994
We already have some great opensource 3D model AI generators, is there a chance that you'd integrate them into Invoke? There are also some ComfyUI workflows that can generate consistent characters, by generating multiple views of the same character then use that data, would be great if you'd integrate that into Invoke too.
@zephilde