For goodness sake, the MET in New York has a massive trove of open CC0 type licensed art. Dear BFL, please ease up a bit on this, and add some art-art to your models, they will be better as a result.
I suspect we'll see the answer to this is LoRAs. Two examples that stick out are:
- Flux Tarot v1 [0]
- Flux Amateur Photography [1]
Both of these do a great job of combining all the benefits of Flux with custom styles that seem to work quite well.
[0] https://huggingface.co/multimodalart/flux-tarot-v1 [1] https://civitai.com/models/652699?modelVersionId=756149
It feels like they just removed names from the datasets to make it worse at recreating famous people and artists.
Diversity is a double-edged sword. It's a desirable feature where you want it, and an undesirable feature everywhere else. If you want an impressionist painting, then it's good to have Monet and Degas in the training corpus. On the other hand, if you want a photograph of water lilies, then it's good to keep Monet out of the training data.
I suspect the same goes for art styles. There's such huge variety that really they'd be better surveys by separate models.
I actually use flux to generate image for purposes of adherence, then pull it in as a canny/depth controlnet with more established models like realvis, unstableXL, etc.
But that real art still exists, and can still be found, so what exactly is the loss here?
It's about 6 lines of Python.
I managed to get it running on an old computer with a 2060 Super, taking ~1.5 minutes per image gen. People are generating on a 1080.
This is miles ahead of most other image generation models available today.
I had similar issues trying to paint a "I cast non-magic missile" meme with a fantasy wizard using a missile launcher. No model out there (I've tried SD, SDXL, FLUX.1dev and now this FLUX1.1pro) knows how a missile launcher looks like (neither as a generic term, nor any specific systems) and even has no clue how it's held, so they all draw really weird contraptions.
flux is amazing, but I find it requires a very literal description, which pushes the "creative work" back to the text itself. Which can certainly be a good thing, just a bit less gratifying to non visual types like myself. :)
I wonder, only somewhat jokingly, if one could make text generators which "imagine" detailed fantastical scenes, suitable for feeding to a text to image model.
"our most advanced and efficient model yet"
"a significant step forward in our mission to empower creators"
I get it, you can't sell things if you don't market them, and you can't make a living making things if you don't sell them, but it's exhausting.
https://huggingface.co/spaces/ArtificialAnalysis/Text-to-Ima...
Some comparisons against DALL-E 3.
- Take your morning to the next level!
What does "state of the art" mean? That it's using the latest "cutting edge" model technology?
When Apple releases a new iPhone Pro Max, it's "state of the art". When they release a new iPhone SE, there's an argument to be made that it's not because it uses 2 year old chips. But what would it even mean for BFL to release a model which wasn't "state of the art"
> our most advanced and efficient model yet
Yes, likewise, this is how technology companies work. They release something and then the next thing they release is more advanced.
> a significant step forward in our mission to empower creators
Going from 12 seconds to 4 seconds is a significant speed boost, but does it move the needle on their mission to empower creators? These are their words, not mine, it's a technical achievement and impressive incremental progress, but are there users out there who are more empowered by this? significantly more empowered!?
There are a lot of things that don't appear in ELO scores. For one, they will not reflect that you cannot prompt women's faces in Flux. We can only speculate why.
I feel like AI should just be treated as fair use as long as its not 100% blatantly a literal clone of the original work.
Ideogram and Flux both have their own broad set of limitations that are non-technical and unpublished. IMO they are not really motivated by legal concerns, other than the lack of transparency itself.
So maybe the issue is that transparency, and that the hazy legal climate means no transparency. You can't go anywhere and see the detailed list of dataset collection and captioning opinions for proprietary models. Open Model Initiative, trying to make a model, did publish their opinions, and they're not getting sued anytime soon. However, their opinions are an endless source of conflict.
edit: nevermind, it's a macos app
When I ask for a man playing accordion, it’s usually a somewhat flawed piano accordion, but If I ask for a woman playing accordion, it’s usually a button accordion. I’ve also seen a few that are half-button, half-piano monstrosities.
Also, if I ask for “someone playing accordion”, it’s always a woman.
No one in the image space wants to admit it, but well over half of your user base wants to generate hardcore NSFW with your models and they mostly don’t care about any other capabilities.
I just tried this Flux1.1 pro page (prompt: "A sad Macintosh user who is upset because his computer can't play games") and was very impressed by the detail and "understanding" this model has.
Ironically, I am afraid to type the website out and will keep it unknown here. My account could be suspended because of this. It had already reached -1 karma. It's better to keep my account alive.
> ... photo with the text "FLUX 1.1 [Pro]", ..., must say "1.1", ...
...And of course, it does not.
