Concept sliders: LoRA adaptors for precise control in diffusion models
sliders.baulab.info
sliders.baulab.info
The main difference is in the training procedure, where they try to only target specific attributes while leaving others unchanged.
You could probably use this to create LoRAs for a specific pose, but if you want to try out many different poses and freely adjust them, the more flexible controlnet approach is likely to be more comfortable.
The distinct thing about concept slider LoRA isn't how much of the image is changed (which will vary between LoRA within the type widely), but that the weight at which the LoRA is applied, rather than just setting how strongly the image tends toward representing a fixed concept, chooses which concept on a continuum the image tends towards.
> What I notice are missing from these examples are actions/poses, so I wonder if it's good for that or we still need openpose/controlnet.
You can do actions/poses via LoRA, but the control you get is qualitatively different than what you get with any of the controlnets, so its good to have both tools available. I haven't seen concepr slider LoRA specifically being used for poses, though conceptually doing something like seated<->standing or standing<->running as a concept slider would make some sense.
You can use SD with image as an input in place of or in addition to a prompt, and you could apply one or more LoRA (concept slider or otherwise) to the model when doing so.
For those in the similar boat, it apparently means Low Rank Adaptation (for Large Language Models)
Well, the abbreviation expands to Low Rank Adaptation, but the article is about diffusion models, not LLMs, and LoRA is an model-modifier approach that applies to many classes of models, not just Large Languag Models.
I wish there was a different description of those who generate art via AI than "artists". When ChatGPT writes me a story based on my prompts I'm still not an "author". Sorry, I know it's a tired argument already but it'll never stop bugging me.
(don't tell my boss)
edit: not my opinion. just prevailing sentiment in yc discussions
I don’t think anyone is arguing to AI personhood yet, so it must be the user who is the artist.
If we want to add extra conditions, that seems fine, but someone else will have to define them. If it were up to me… I’d probably end up being called a gatekeeper.
For a counterexample: suppose you're really good at sketching, and you go to the Louvre and copy out the Mona Lisa. Were you the sole artist of your sketch? Leonardo Da Vinci certainly wasn't drawing on your notebook.
In the case of AI Art, there is no point in the process where you are participating in the creation of the actual initial image unless the actual source/model is based of your own artwork that you created. What you are really doing with AI art is generating a concept based on a request you make (the prompt), and then modifying the result with other tools until you get what you want (inpainting masks, animate it, etc.).
The equivalent of this would be something of like a patron/commissioner and something like a sound engineer. You aren't creating the source material, but you can ask an AI art generator to create something based on your request, and then modify and tune it afterwards to your liking. It is more of like a producer/sound engineer/post-production role.
The "Desired Look" can be considered the "artistic" part of art, while the production can be considered the craft.
Koons is the most extreme example I can think of, but there are plenty of others who got their start creating stuff themselves and then hired people to do the total of the work.
Is a director an artist, when he doesn't write the script, touch the camera, edit the film, or act in the movie?
Ironically the photography community already went through this sort of discussion as the Art of the darkroom became the 'art' of photoshop
Though the scale is lessened due to the significant barrier of entry for high end photography (a few grand per body and per lens)
Just look at ComfyUI: https://github.com/comfyanonymous/ComfyUI
Regardless, the original comment was clearly not serious.
I don't get paid enough to be a titan.
It certainly has its place, but the "its taking people's jobs" argument kinda fell apart for me after that moment
AI can be used as a tool to empower authors and artists to create more and/or better works, but it can also be used by a lazy student to finish their 2 page writing assignment in seconds.
The only thing AI will clearly do over the next few years is raise the bar. Beginner art or copywriting is already outclassed and assignments will assume access to AI tools, similarly how many assume internet access nowadays.
Nevertheless, I truly believe that there will always be space for human art... just look at chess, with engines far better than humans, what are the tournaments about? the players.
That does not make you an artist whenever you spill ketchup on your shirt, wrap a Chrismas gift or pour yourself a scotch on the rocks.
Second, while you can indeed just type a prompt into Midjourney and some images will be generated, there are 'artists' using diffusion models to meticulously craft 'art' using hundreds of itrrations combining dozens of models and loras with inpainting. outpainting, recomposing and fusing, dtawing and more to come to a unique piece of 'art'.
Third, didn't we have the same thing before? Are DJ's musicians? Some most definetly are. Creating new music by adapting, restructuring. remixing and extending existing tracks, while others use turntables as a real instrument. Others just put on some records at a wedding.
It's about intention and context, and yes, creativity and skill, not about the tools used.
I would also add that art is also "created" when you consume it. When we consume it, we generate some interpretation, feelings or imagery. This is the last step. That's internal, but this processing of art is also art.
That explain why some people find something as art and others as BS. As you say it's about intention and context, from the artist point of view and also from the consumer.
AI "artist" isn't an artist. It's someone who uses an external product to automate the talent for them by creating an average image based on samples created by artists, of which were curated by individuals with particular tastes. It's why AI art mostly looks the same.
It's also the reason a DJ a is called a disc jockey and not a musician. They are labeled differently to show the difference in the generation of the supplied content.
You might as well say that someone who Googles "picture of a dog" is an artist as well since they learned the ins and outs of the search engine. A better term for AI Artist would be something that mentions the collage aspect of it.
Typically, if you generate art, regardless of the tool or medium or form, you are an artist.
There may also be a more specific term for artists working with/in a specific tool, medium, or form, but the general term still applies.
Stable Diffusion is a tool used yb a person, not a person.
It's also not a tired argument. You are just posting in a place that is very pro tech and not art-oriented. If you posted this in an art forum you'd get the opposite reaction.