I've been using openjourney (and MJ/SD) quite a bit, and it does generate "better" with "less" compared to standard v1.5, but it's nowhere close to Midjourney v4.
Midjourney is so far ahead in generating "good" images across a wide space of styles and subjects using very little prompting. While SD requires careful negative prompts and extravagant prompting to generate something decent.
Very interested in being wrong about this, there's so much happening with SD that it's hard to keep up with what's working best.
I've been thinking that for months, but recently swung towards being more optimistic about SD again, everything midjourney looks midjourney while SD allows you to create images in any style. MJ really needs to get rid of that MJ style, make it optional as it's undeniably pretty, it's just becoming a little much.
But I still feel 2.x is somehow a degradation of 1.x, its hard to get something decent out of it. The custom training/tuning and all is nice (and certainly the top rain to use SD over MJ, many use cases MJ just can't do) but it should not be used as a band-aid for appearantly inherent shortcomings in the new clip layer (I'm assuming this is where the largest difference comes from, since the Unet is trained on largely the same dataset as 1.x).
To be fair that's the default style of MJ, you're seeing that a lot because most users don't take the time to add style modifiers to their prompts.
If you add qualifiers such as soft colors, impressionistic, western animation, stencil, etc. you can steer mid journey towards much more personalized styles.
Yeah, a lot of Midjourney images are very clearly Midjourney images. Does Midjourney have inpainting/outpainting yet? I admit it's the offering I've evaluated the least.
Midjourneys upscaled images to their current max offering look fantastic, that's for sure. My wife generates some really great stuff just for fun.
I think 2.0 has potential still, it works with textual inversion type models much better, which can kinda play nice with each other, so given enough of those I imagine you can get some cool stuff with it. I've also heard it does negative prompts much better, so those are less optional in 2.0
But yeah for now, all my custom models are 1.5 so I've yet to fully upgrade yet, most of the community seems to be doing similar at the moment.
Every midjourney image has the same feeling to it. A bit 1950s sci-fi artist. I guess it's just that it all looks airbrushed? I can't put my finger on it.
I don't know if that was Midjourney's intent, but it seems like a smart approach. Instead of trying to be everything to everyone and generating quite a lot of ugly garbage, you get consistently good-looking stuff in a certain style. I'm sure it helps their business model.
Feels like it's the Instagram model for prompt-generated images.
Anyone can get a camera phone, take a picture and use some free software (e.g. gimp) to get great results in post-processing.
Most non-expert users though want to click on a few pre-defined filters, find one they like & run with it, rather than having more control yet poorer results (precisely because they _aren't_ experts).
If Midjourney applies this to all their artwork then maybe it alleviates some of the ethical concerns (Midjourney then has a "style" independent of the training data)
I've played a lot with it lately and that just not true. If you play with styles, colors, angles, views you have a lot of control about how the imagine will look. It can emulate pretty much all mainstream aesthetics.
This is just a sd checkpoint trained on output of Midjourney. You can load it into a1111 or invokeai for easier usage. If you are looking for new checkpoints, check out the Protogen series though for some really neat stuff.
Not sure why this is downvoted. Civitai does in fact list a bunch of fine tuned models and can be sorted by highest ranked, liked, downloaded, etc. It is a good resource. Many of the models are also available in the .safetensor format so you dont have to worry about a pickled checkpoint.
I didn't downvote but I have to say the images shown on that page are hilariously juvenile. I was a teenager once so I get it but I'm guessing the content is where the downvotes are coming from?
Looking at this site, I would argue that the canonical "hello world" of an image diffusion model is a picture of a pretty woman. The canonical "hello world" for community chatbots that can run on a consumer GPU will undoubtedly be an AI girlfriend.
BTW I love your app! At my desk I use Automatic1111 (because I have a decent GPU), but it's so nice to have a lean back experience on my iPad. Also, even my 6yo son can use it, as he doesn't need to manipulate a mouse.
I just gave Protogen a spin and the diversity of outputs it gave me was abysmal. Every seed for the same (relatively open-ended) prompt used the same color scheme, had the same framing, and the same composition. Whereas with SD 1.5/2.1, the subject would be placed differently in-frame, color schemes were far more varied, and results were far more interesting compositionally. (This is with identical settings between the two models and a random seed)
So unless you want cliche-as-fuck fantasy and samey waifu material, classic SD seems to do a much better job.
Yes, protogen is based on merging of checkpoints. The checkpoints it's merged from are also mostly based on merging. Tracing the degree of ancestry back to fine tuned models is hard, but there's a ton of booru-tagged anime and porn in there.
If there's one style I dislike more than the bland Midjourney style, it's the super-smooth "realistic" child faces on adult bodies that protogen (and its own many descendants) spit out.
Mj earlier versions were around before SD came out. Before dall-e 2 too, but after 1 IIRC. So I assume they have their own custom setup. Perhaps based on dall-e 1 paper originally (not weights as they were never published) and improved from there.
Interesting i thought stable diffusion was the only other "big player" besides OpenAI because of the expenses in training and extrapolating from papers / new research.
Is Midjourney heavily funded? Because if they can battle SD why aren't we seeing lots of people doing the same, even in the Open Source space?
sd checkpoint -- stable diffusion checkpoint. a model weights file that was obtained by tuning the stablediffusion weights file using probably something like dreambooth on some number of midjourney-generated images.
a1111 / invokeai -- stable diffusion UI tools
Protogen series -- popular stablediffusion checkpoints you can download so you can generate content in various styles
> This is just a sd checkpoint trained on output of Midjourney
Which is sub-optimal -> bad. You don't want to train on output from an AI because you'll end up with a worse version of whatever that AI is already being bad at (hands, foot, and countless other things). This is the AI feedback loop that people have been talking about.
So instead of figuring out what Midjourney has done to get such good result, people just blatantly straight copied those results and fed them directly into the AI, as true as the art thief stereotype they are.
The huggingface element of these annoys me. Reading the other comments, this is just a stable diffusion checkpoint, so I should be able to download it and not use the diffusers library or whatever other HF stuff. But it's frustrating that it's tied to a for profit ecosystem like this.
I suppose pytorch is / was Facebook, but if feels more arms length. I don't have to install and run a facebook cli to use it (nobody get any ideas).
You don't need a HF cli, you just need to use git LFS (I believe now part of git) to pull the files off of HF (unfortunately still requiring an account with them). It would be nice to see truly open mirrors for this stuff that don't have to involve any company.
Is it possible to download with curl or git lfs (or other "free" command line tool) with no login? I couldn't find a way to do that with the original sd checkpoints.
i don't think it's at the point where most individuals can financially support the model training, its a company doing all this because it requires the consolidated funds of a business
You can download the checkpoint right from hugging face and diffusers is a library you can use for free, I'm not sure what the issue is here, that people need an account?
Q: "Are OpenRAILs considered open source licenses according to the Open Source Definition? NO."
A: "THESE ARE NOT OPEN SOURCE LICENSES, based on the definition used by Open Source Initiative, because it has some restrictions on the use of the licensed AI artifact.
That said, we consider OpenRAIL licenses to be “open”. OpenRAIL enables reuse, distribution, commercialization, and adaptation as long as the artifact is not being applied for use-cases that have been restricted.
Our main aim is not to evangelize what is open and what is not but rather to focus on the intersection between open and responsible licensing."
FWIW, there's a lot of active discussion in this space, and it could be the case that e.g. communities settle on releasing code under OSI-approved licenses and models/artifacts under lowercase "open" but use-restricted licenses.
My biggest critique of OpenRAIL is that it's not entirely clear that AI is copyrightable[0] to begin with. Specifically the model weights are just a mechanical derivation of training set data. Putting aside the "does it infringe[1]" question, there is zero creativity in the training process. All the creativity is either in the source images or the training code. AI companies scrape source images off the Internet without permission, so they cannot use the source images to enforce OpenRAIL. And while they would own the training code, nobody is releasing training code[2], so OpenRAIL wouldn't apply there.
So I do not understand how the resulting model weights are a subject of copyright at all, given that the US has firmly rejected the concept of "sweat of the brow" as a copyrightability standard. Maybe in the EU you could claim database rights over the training set you collected. But the US refuses to enforce those either.
[0] I'm not talking about "is AI art copyrightable" - my personal argument would be that the user feeding it prompts or specifying inpainting masks is enough human involvement to make it copyrightable.
The Copyright Office's refusal to register AI-generated works has been, so far, purely limited to people trying to claim Midjourney as a coauthor. They are not looking over your work with a fine-toothed comb and rejecting any submissions that have badly-painted hands.
[1] I personally think AI training is fair use, but a court will need to decide that. Furthermore, fair use training would not include fair use for selling access to the AI or its output.
[2] The few bits of training code I can find are all licensed under OSI/FSF approved licenses or using libraries under such licenses.
Not a lawyer, but as I understand the most likely way this question will be answered (for practical purposes in the US) is via the ongoing lawsuits against GitHub Copilot and Stable Diffusion and Midjourney.
I personally agree the creativity is in the source images and the training code, but think that unless it is decided that for legal purposes "AI Artifacts" (the files containing model weights, embedding, etc.) are just transformations of training data and therefore content and subject to the same legal standards as content, I see a lot of value in trying to let people license training and code and models separately. And if models are just transformations of content, I expect we can adjust the norms around licensing to achieve similar outcomes (i.e., trying to balance open sharing with some degree of creator-defined use restriction).
The co-pilot and Dalle lawsuits aren't about if the training weights file can be copyrighted though (they are about if people's work can be freely used for training).
This is a different issue where the OP is arguing that the weights file is not eligible for copyright in the US. That's an interesting and separate point which I haven't really seen addressed before.
The two issues aren't exactly the same but they do seem intimately connected. When you consider what's involved in generating a weights file, it's a mostly mechanical process. You write a model, gather some data, and then train. Maybe the design of the model is patentable, or the model/training code is copyrightable (actually, I'm pretty sure it is), but the training process itself is just the execution of a program on some data. You can argue that what that program is doing is simply compiling a collection of facts, which means you haven't created a derivative work, but in that case the weights file is a database, by definition, so not copyrightable in the US. Or you can argue that the program is a tool which you're using to create a new copyrightable work. But in that case it's probably a derivative work.
Appreciate the distinction in the above comment that they are two distinct questions, but also agree the two questions are very connected.
I should've been more specific: I was thinking mainly of the artists v. stable diffusion lawsuit which makes the specific technical claim that the stable diffusion software (which includes a bunch of "weights files") includes compressed copies of the training data. (Line 17, "By training Stable Diffusion on the Training Images, Stability caused those images to be stored at and incorporated into Stable Diffusion as compressed copies", https://stablediffusionlitigation.com/pdf/00201/1-1-stable-d...).
I expect that if the decision hinges on this claim, that could have far reaching implications re: model licensing. I think this along the lines of what you've laid out here!
How would you distinguish “just a mechanical derivation of training set data” from compiled binary software? The latter seems also to be a mechanical derivation from the source code, but inherits the same protections under copyright law.
Usually binaries are compiled from your own source code. If I took leaked Windows NT kernel source and compiled it myself, I wouldn't be able to claim ownership over the binaries.
Likewise if I drew my own art and used it as sample data for a completely trained-from-scractch art generator, I would own the result. The key problem is that, because AI companies are not licensing their data, there isn't any creativity that they own for them to assert copyright over. Even if AI training itself is fair use, they still own nothing.
Do artists not own copyright on artwork which comprises other sources (eg. collage, sampled music)? It’d be hard to claim that eg. Daft Punk doesn’t own copyright on their music.
(Whether other artists can claim copyright over some recognisable sample is another question.)
This is why there's the "thin copyright" doctrine in the US. It comes up often in music cases, since a lot of pop music is trying to do the same thing. You can take a bunch of uncopyrightable elements, mix them together in a creative way, and get copyright over that. But that's a very "thin" copyright since the creativity is less.
I don't think thin copyright would apply to AI model weights, since those are trained entirely by an automated process. Hyperparameters are selected primarily for functionality and not creative merit. And the actual model architectures themselves would be the subject of patents, not copyright; since they're ideas, not expressions of an idea.
Related note: have we seen someone try to patent-troll AI yet?
The Verve's Richard Ashcroft lost partial copyright and all royalties for "Bitter Sweet Symphony" because a sample from the Rolling Stones wasn't properly cleared: https://en.m.wikipedia.org/wiki/Bitter_Sweet_Symphony
Men at Work lost copyright over their famous "Land Down Under" because it used a tune from "Kookaburra sits in the Old Gum Tree" as an important part of the chorus.
>Do artists not own copyright on artwork which comprises other sources (eg. collage, sampled music)? It’d be hard to claim that eg. Daft Punk doesn’t own copyright on their music.
“Mechanical derivation” is doing a lot of heavy lifting here. What qualifies something as “mechanical”? Any algorithm? Or just digital algorithms? Any process entirely governed by the laws of physics?
So, in the US, the bedrock of copyrightability is creativity. The opposite would be what SCOTUS derided as the "sweat of the brow" doctrine, where merely "working hard" would give you copyright over the result. No court in the US will actually accept a sweat of the brow argument, of course, because there's Supreme Court precedent against it.
This is why you can't copyright maps[0], and why scans of public domain artwork are automatically public domain[1][2]. Because there's no creativity in them.
The courts do not oppose the use of algorithms or mechanical tools in art. If I draw something in Photoshop, I still own it. Using, say, a blur or contrast filter does not reduce the creativity of the underlying art, because there's still an artist deciding what filters to use, how to control them, et cetera.
That doesn't apply for AI training. The controls that we do have for AI are hyperparameters and training set data. Hyperparameters are not themselves creative inputs; they are selected by trial and error to get the best result. And training set data can be creative, but the specific AI we are talking about was trained purely on scraped images from the Internet, which the creator does not own. So you have a machine that is being fed no creativity, and thus will produce no creativity, so the courts will reject claims to ownership over it.
[0] Trap streets ARE copyrightable, though. This is why you'll find fake streets that don't exist on your maps sometimes.
[2] Several museums continue to argue the opposite - i.e. that scanning a public domain work creates a new copyright on the scan. They even tried to harass the Wikimedia Foundation over it: https://en.wikipedia.org/wiki/National_Portrait_Gallery_and_...
Possibly, but even if that were the case, it would protect NovelAI, not Stability.
The closest analogue I can think of would be copyrighting a Magic: The Gathering deck. Robert Hovden did that[0], and somehow convinced the Copyright Office to go along with it. As far as I can tell this never actually got court-tested, though. You can get a thin copyright on arrangements of other works you don't own, but a critical wrinkle in that is that an MTG deck is not merely "an arrangement of aesthetically pleasing card art". The cards are picked because of their gameplay value, specifically to min-max a particular win condition. They are not arrangements, but strategies.
Here's the thing: there is no copyright in game rules[1]. Those are ideas, which you have to patent[2]. And to the extent that an idea and an expression of that idea are inseparable, the idea part makes the whole uncopyrightable. This is known as the merger doctrine. So you can't copyright an MtG deck that would give you de-facto ownership over a particular game strategy.
So, applying that logic back to the training set, you'd only have ownership insamuch as your training set was selected for a particular artistic result, and not just "reducing the loss function" or "scoring higher on a double-blind image preference test".
As far as I'm aware, there are companies that do creatively select training set inputs; i.e. NovelAI. However, most of the "generalist" AI art generators, such as Stable Diffusion, Craiyon, or DALL-E, were trained on crawled data without much or any tweaking of the inputs[3]. A lot of them have overfit text prompts, because the people training them didn't even filter for duplicate images. You can also specifically fine-tune an existing model to achieve a particular result, which would be a creative process if you could demonstrate that you picked all the images yourself.
But all of that only applies to the training set list itself; the actual training is still noncreative. The creativity has to flow through to the trained model. There's one problem with that, though: if it turns out that AI training for art generators is not fair use, then your copyright over the model dissolves like cotton candy in water. This is because without a fair use argument, the model is just a derivative work of the training set images, and you do not own unlicensed derivative works[4].
[1] Which is also why Cory Doctorow thinks the D&D OGL (either version) is a water sandwich that just takes away your fair use rights.
[2] WotC actually did patent specific parts of MTG, like turning cards to indicate that they've been used up that turn.
[3] I may have posted another comment in this thread claiming that training sets are kept hidden. I had a brain fart, they all pull from LAION and Common Crawl.
[4] This is also why people sell T-shirts with stolen fanart on it. The artists who drew the stolen art own nothing and cannot sue. The original creator of that art can sue, but more often than not they don't.
(The format of the blob is python's "pickle" format: a binary serialization of an in-memory object, used to store an in-memory object and later load it, perhaps on a different machine.)
But, I did not find any source code for generating that file. Am I missing something?
Shouldn't there at least be a list of input images, etc and some script that uses them to train the model?
Hahahahaha you sweet summer child. Training code? For an art generator?!
Yeah, no. Nobody in the AI community actually provides training code. If you want to train from scratch you'll need to understand what their model architecture is, collect your own dataset, and write your own training loop.
The closest I've come across is code for training an unconditional U-Net; those just take an image and denoise/draw it. CLIP also has its own training code - though everyone just seems to use OpenAI CLIP[0]. You'll need to figure out how to write a Diffusers pipeline that lets you combine CLIP and a U-Net together, and then alter the U-Net training code to feed CLIP vectors into the model, etc. Stable Diffusion also uses a Variational Autoencoder in front of the U-Net to get higher resolution and training performance, which I've yet to figure out how to train.
The blob you are looking at is the actual model weights. For you see, AI is proprietary software's final form. Software so proprietary that not even the creators are allowed to see the source code. Because there is no source code. Just piles and piles of linear algebra, nonlinear activation functions, and calculus.
For the record, I am trying to train-from-scratch an image generator using public domain data sources[1]. It is not going well: after adding more images it seems to have gotten significantly dumber, with or without a from-scratch trained CLIP.
[0] I think Google Imagen is using BERT actually
[1] Specifically, the PD-Art-old-100 category on Wikimedia Commons.
The SD training set is available and the exact settings are described in reasonable details:
> The model is trained from scratch 550k steps at resolution 256x256 on a subset of LAION-5B filtered for explicit pornographic material, using the LAION-NSFW classifier with punsafe=0.1 and an aesthetic score >= 4.5. Then it is further trained for 850k steps at resolution 512x512 on the same dataset on images with resolution >= 512x512.
To my eye, kmeisthax's comment appears to be entirely accurate.
Well, that is to say, that assuming the facts that listed are accurate, then I agree with the conclusion: it's not "open source" at all. (And certainly not Libre.)
The things you said do not describe an open source project.
The point here is that the title of this thing is incorrect. If the ML community doesn't agree, it's because they are (apparently) walking around with incorrect definitions of "open source" and "Free Software" and "Libre Software".
Seems like 'the lawyers who made the license' and the OSI might be good authorities on what's open source. I'd love to hear a good FSF rant about RAIL though.
Well Open Source licenses don't make sense for training artifacts for the same reason Creative Commons licenses are used for written and artists "open" works rather than Open Source.
> Also, you can make a carrier! How you may ask? it is easy. In our time, we have a lot of digital asset marketplaces such as NFT marketplaces that you can sell your items and make a carrier. Never underestimate the power open source software provides.
At first I thought this might be a joke site, the poorly written copy reads like a parody.
Also, as others have pointed out, this is basically just yet another Stable Diffusion checkpoint.
This particular wording sounds like it could be a poor translation from Russian. Sdelat' karjeru (literally: to make a career) means to make a living doing something, or to succeed in doing some job.
I was about to integrate this into https://88stacks.com but it requires a write token to hugging face which makes no sense. It’s a model that you download. Why does it need write access to hugging Face!?!
Does it really, have you tried it or do you mean because of the documentation? Just skimmed through the code, haven't really seen anything related to uploading. Might not even be required.
Maybe this is an obvious question but if you generate pictures using any of these tools, can you create the same picture/character/person with different poses, or backgrounds, such as for telling a story, and/or creating a comic book, or would you get a new picture every time, such as for the cover of a magazine ?
Yes, you can create an AI model based on a few pictures of the “model” (the model can also be AI generated) and then you can generate images of all kinds with that model included.
This was my project, but since @prompthero changed their "midjourney-v4 dreambooth" model's name to openjourney, I changed my model name to "Mann-E" which is accessible here:
https://huggingface.co/mann-e/mann-e_4_rev-0-1
(It's only a checkpoint and under development)
I'm failing to train a model off of this in the Automatic1111 webui Dreambooth extension. Training on vanilla 1.5 works fine. It throws a bunch of errors I don't have in front of me on my phone.
Anyone else have similar issues? I loaded it both from a locally downloaded version of the model as well as from inputting in the huggingface path and my token with write (?!?) permissions.
Who knows what Midjourney uses. We've got only claims in discords to go by.
My guess is they do internally a slightly more careful and less porn/anime oriented version of what the 4chan/protogen people do. Make lots of fine tuned checkpoints, merge them, fine tune on a selection of outputs from that, merge more, throw away most of it, try again etc. Maybe there are other models in the mix, but I wouldn't bet on it.
noob question - how hard is it to setup and run this on a windows machine? I've had bad luck with python and package management in windows in the past but that was a long time ago.
If you use the webui it's a single git clone and an optional file edit to set some CLI flags and that's it. You download models and move them to a directory to use them. Recently they introduced a binary release for people that are unfamiliar with git.
I've only tried out stable diffusion to any real extent, but seeing what other people have gotten out of the other two I can easily say it's the least performant of the bunch.
I would be hesitant to pass judgement if only playing with one. It's easy to compare the deluge you've picked through to other people's best picked cherries...
Well sure, but after hours and hours of messing with params my cherry picked best cases were still lightyears away from the average Midjourney example. Maybe I'm just bad at it ¯\_(ツ)_/¯
Looks good but this works well only for gamey, sci-fi kind of themes. Any suggestions for prompts which can yield interesting flowcharts to explain technical concepts?
This is just openjourney model fine-tuned with Dreambooth. You can use any of these tools: Draw Things, Mochi Diffusion, DiffusionBee, AUTOMATIC1111 UI on M1 / M2 with this model. (I wrote Draw Things).
Ought to! There are some enhancements coming down the pipe for Macs w/ CoreML, so while they won't be as fast as having a higher end NVidia, they'll continue to get performance increases, as well.
Some self promotion.
We got Stable Diffusion made available as SaaS on AWS[1] with per minute pricing and the unique thing with our SaaS offering is you can shutdown/restart the SaaS environment yourself . You will get charged on per minute basis only when the environment is running.
Also, if you want to try the SaaS for free, feel free to submit a request using our contact-us form [2]
I've been using openjourney (and MJ/SD) quite a bit, and it does generate "better" with "less" compared to standard v1.5, but it's nowhere close to Midjourney v4.
Midjourney is so far ahead in generating "good" images across a wide space of styles and subjects using very little prompting. While SD requires careful negative prompts and extravagant prompting to generate something decent.
Very interested in being wrong about this, there's so much happening with SD that it's hard to keep up with what's working best.