not-for-all-audiences glitch is normally fixed
Hello @Yntec , just to prevent you that the not-for-all-audiences tag appearing in some models aswell that one, but was wrongly appearing, is now fixed normally!
Hey FashionStash! The bug I'm fighting is that if I move the not-for-all-audiences tag position so it looks like in this model: https://huggingface.co/John6666/noobai-xl-nai-xl-earlyaccessversion-sdxl then the model can't be used in a space like this: https://huggingface.co/spaces/Yntec/ToyWorldXL
The fight was for being able to test the noobai/illustrious models but after I managed to test them they disappointed me so I didn't push for featuring other models (I could get ToyWorldXL to 400 models, but there's clearly not much interest in this, while SD1.5's interest remains strong, so I let people vote with their usage of the models I upload - it couldn't get to 400 downloads, HiFi got to 1K in a day, so I'll keep focusing in the kind of models people use.)
@Yntec ahemmmm... In fact, I recall each times I tested SDXL models, it was amazingly realistic, but also poorly creative, so SDXL not my best friends when I need more creative than photo realism 😂!
And about the Not-For-All-Audiences tag, I think it's all related to the NSFW settings in each people account, and then can only set-up in a private model or private space.
I just suppose!
Currently I'm trying funny "breakfasts" with Lah-dee-da, the result is not a good thing for yet, I must be more carefully precise in the prompt.
I only use basic settings of Inference-api serverless, it's more what I can call by letting a model being creative by itself 😂!
By the way I need retrieve some of my older prompts I tested in Gacha but to get them working on.... LadyNostalgia!
I'm afraid the Not-For-All-Audiences models may be good at that but not at the things I usually prompt! I wouldn't know because I normally don't test for those...
Currently I'm trying funny "breakfasts" with Lah-dee-da, the result is not a good thing for yet, I must be more carefully precise in the prompt.
Cool! If you could tell me your prompt and an example image (maybe from google images?) of what you wanted it to look like, perhaps I could make a model good at that! Or perhaps not! I never tried it but it's an interesting concept to experiment with.
By the way I need retrieve some of my older prompts I tested in Gacha but to get them working on.... LadyNostalgia!
I'm looking forward to your results and being thankful for your feedback!
NFAA tag is a good thing.
It's reducing the pain to moderate models. But false detection is such of an hard pain when it hits Illustrious models.
Censorship is never a good thing, it's always limiting creativity behind wrong filters detection (how many times online ChatGPT flagged my discussions as inappropriate when I added the trigger term without any idea where it flagged my discussion as inappropriate, even for StableDiffusionWeb.com).
Censorship is always bad.
But it is always the last response to regulate strongly the things that must be regulated.
The issue here with backend censorship NFAA tag, it's that it make a model totally unable to run on HuggingFace Inference-api (paid or free or from an endpoints). So too much sadistic to see a model locked behind a wrong tag!
.......
For other things :
Hey sure! Will give that next week on your La-dee-dah model's page my "breakfast" concept art prompts and images ! 😁
With your models which I love so much more than the original ones (I find them so much creative, LadyNostalgia (Gacha, your PhotoReal2. 0 upgrade) are full lovely proves of what I said!)).. I essentially prompt arts and concepts arts.
Which serves (for the best images for illustrating covers/graphics for my Wattpad stories. And also for getting a huge amount of Trading Card Games illustrations for physical tabletop games I create sometimes, just for myself and my family and my friends in real life.
I can give you also those one!
You're definitively right about censorship, but I'm glad there's a workaround, if you want to use any of the Illustrious models tell me so I add them to https://huggingface.co/spaces/Yntec/ToyWorldXL where they can be used fine, most of the models I want to use have their Inference API off anyway due to low activity so I have to rely on my spaces to use them, and then we have Stable Diffusion 3.5 Medium which is active all the time, but hasn't worked for the last two days so I've been unable to use it there! (I was about to add it to ToyWorld, what a bad timing, but at least it's not my model...)
Which serves (for the best images for illustrating covers/graphics for my Wattpad stories. And also for getting a huge amount of Trading Card Games illustrations
I'm glad to hear that! Art is a difficult subject, my most successful models have been either for Anime or photorealism, but IsThisArt was the only one that hit thanks to the beautiful faces it draws for girls.
I have made Stable Diffusion 1.5 my comfort zone and almost never venture outside of it, recently I did it for the first time with AnotherFineMess, just to have a model with a different style and compositions from all others, but it suffers heavily on the coherency and anatomy department, I don't think you'd want to use it for trading card game graphics unless you'd want to create monsters with 3 arms or 6 fingers... but it got 20000 downloads! I never expected that, I've uploaded much better models that never got to 1/10th of that, but it seems model uniqueness pays and a different architecture is the cheapest way to get it.
I can give you also those one!
Yes! That sounds great! I'll be merging La-dee-dah with more models once I know what should I be aiming for to deliver, I'm looking forward to it and it's always a pleasure talking with you!
Hello @Yntec , It's always a pleasure to me to discuss with you!
I will not able to continue our discussion (here) as when I wanted answers, you was gone and that page (discussion) was no more accessible at all!
But you're back, that's very good news (at least for me, as I said I appreciate you !)
Well by what starting my answer.. Humm...
Ok
I see you doed some new fancy models!
I will try it out when I get time to go to the test road!Precisely for the moment, I test mainly SD 1.5 models.
(currently, as you came back from the void, I'm tied up with lot of fun in continuing to get surprised by Gacha and LadyNostalgia!).I'm even not tested Stable Diffusion 3.5 or SDXL Illustrious (I just wanted to test it, but I am addicted to your SD 1.5 models! 🤭 Which are the best for me in SD 1.5 conversions (and as previously said, on original!).
Also, SDXL models tests from free HuggingFace Inference-api goes much too often in MODEL TIME-OUT, impossible to working at best to not waste time with them.
Your new models are integrating some Paintings logics which is really interesting!
You said AnotherFineMess, got 20k downloads? That's incredibly impressive! You know.. Sometimes prompts can get a "seems-to-be-bad" model into an "incredible" model. I mean, sometimes, prompts when carefully written or organized are very improving models quality for rendering final images!
I will show you many pictures I generated with model I already tested and loved in your amazing models collection!
Hey @FashionStash , we hadn't talked in a while!
I will not able to continue our discussion (here) as when I wanted answers, you was gone and that page (discussion) was no more accessible at all!
Whoops! That was not intentional, I destroyed many things in the process and restoring them isn't trivial, but I have prioritized that one so it's now back online!
https://huggingface.co/spaces/Yntec/Image-Models-Test-September-2024/discussions/1
I test mainly SD 1.5 models.
I keep focusing on them, mainly because the rest of the technologies are so heavy I can't really do my thing with them! But future models have suffered from what has been called the "AI slop", there's something off about them, not about realism or quality, it's the same thing that is off about stock photos, they don't feel real even if real people are in them, so what can ne expect from AI generations? Apparently they have maximized "aesthetics", so I'll say I don't like aesthetics, some of my merges actually get rid of aesthetics on purpose...
I just wanted to test it, but I am addicted to your SD 1.5 models!
As I'm addicted to making them! As always I keep getting more models in the plans than the ones I deliver, a problem is it's very difficult to provide something better, like the AnalogWorld model I just released, it's not really better than my Analog model that I had. Analog has better coherence and compositions, and it's easy to mimic AnalogWorld's color by modifying the prompt. Oh well! I'll just get used to the idea of providing worse models than before, one can't just go and make something better so easily.
too often in MODEL TIME-OUT
Yeah, after https://huggingface.co/spaces/stabilityai/stable-diffusion-3.5-medium went offline I haven't used SD3.5 Medium even once, it just keeps timing out in the serverless API. LEOSAM did a great work bringing the SDXL style to SD1.5, I hope he just does some FilmGirl version with SD3.5 Medium's style.
Your new models are integrating some Paintings logics which is really interesting!
Thanks! There's many concepts I have not explored yet because I don't feel I'm ready yet. And sometimes I wonder if I'll ever be ready, but I liked the results of these ones, juaner0211589 makes great Loras.
got 20k downloads? That's incredibly impressive!
Yeah, I learned people value variety since for SD1.5 you can always find another model that is kind of similar, but AnotherFineMess looks very different from all other SD1.5 models.
I will show you many pictures I generated with model I already tested and loved in your amazing models collection!
I'll be looking forward to that!