Stable Core

3. A third prompt at least produced a publishable result. However, I tried this with different models and other images looked better. 4. I am concerned that some recently published models have limited setting options for the user. It may well be that many users like it when the AI does the main work. My motivation for being with NC and not somewhere else is that I have more setting options here (start image, inpainting). I would therefore like to give constructive feedback that I would like to see future development move in the direction of retaining and further developing models with setting options. This is an added value that NC has over other platforms.

I'll second that. And it's very disappointing to see so many fantastic images and when I check the prompt (if it's open) It says "a cat". I would be terribly ashamed to publish that. A yet there are many that do. Like, look at my world class masterpiece "I" created! Come check out all my amazing 1 word prompt masterpieces! Makes me wonder how many hidden prompts are hidden because of just that.🤔 To much reliance on automation. I want to see creative artwork directed by humans!

Thank you very much for agreeing! Yes, I see it the same way. For me, art is always an expression of another person's soul. I want to see other people's ideas, thoughts and feelings condensed into a work of art in order to understand each other better. I also like beautiful pictures and find AI very exciting. But the first thought is more important to me personally. Therefore, I would like it if the AI is primarily a tool like a paintbrush, but the idea comes from a human who has experienced joy and suffering and wants to communicate something to me. And who has creative ideas to get something out of the AI that is not just part of the main training.

Could not agree more. Well said buddy.

You've got a good point there. SDXL lightning has a lower price, so less features is normal. Stable Core also just doesn't support start images. Stable Diffusion 3 will though, and there are separate models for editing images (basically, inpainting, outpainting) So, all these features are still here in new models. The problem is whether it's able to be implemented here in NC. It would be a shame if these functions were removed in order to make it more cost effective.

I agree with you, we would like more flexibility added to the setting options in all models, also believe like you that this model is too expensive to use and the results barely compare with the amount of detail that DreamShaper XL Lightning provides, and I also wish that DS XL lightning as well as this model would allow to do inpaint or progress a starting image. Like you, I also feel that my motivation drops when we find these roadblocks.

I'm glad that you agree with me and that I'm not alone in my opinion. This encourages me to pass this feedback on to the NC team. I think it's important that the team knows the interests of different users.

I do agree that art is an expression of the soul. However, I do understand that it's difficult to engineer a good prompt. A lot of time and energy is required into creating a good prompt, and not everyone has the time (and credits) to do so. However, I do notice that the models are gradually becoming "easier" to use. Before SDXL 0.9, most of us were able to get something good out of SDXL beta, or even SD1.5. Ever since the "good" models were on NC, we would just transfer our existing skills to the new model. However, this skill is not a necessity anymore; new users won't even need to learn how to engineer a prompt anymore, and still come up with good-looking results. It's probably just human nature, many of us will take the easier path.

For me, there is nothing wrong with models being easy to use and producing beautiful images with little effort. However, that does not mean that existing features should disappear. Things like start image, evolving, multiple prompts, prompt weights or inpainting are options that no one needs to know about or use. But it would be nice if you could still use them. That's what I wanted to express with point 4: some of the newer models no longer support all of these options. I am therefore concerned that we will reach the point where these options will be abolished. I would simply find that a shame because I think they are valuable for creative variation and technical experimentation.

Hey man, back to Stable Core not being any real improvement at all. Have you taken Stable Diffusion 3 for a spin yet? If not you don't know what you're missing! I've been using the BETA for a while. Some of my results are up on my gallery along with their prompts. I've had a few come out with lackluster visuals, and if you look closely there are some typical flaws yet to be worked out, but the results have been unbelievable! In the tiger scene I asked for 15 specific things without weighting anything, and got EVERY single one of them! Including a tiger, macaw and a toucan. Along with a waterfall, the tiger being where I asked he be, some white clouds, moss, deadfall... When it's ready it's going to make SDXL into a dinosaur. Also I haven't had a single case of specified colors "bleed", which still happens with Dall. Crazy dude.

☝️🤓 I opened a club house.

That's a nice idea, Knusperfunk. I'm joining your room. 🙂

Many of the newer models don't have these features yet.

ChatGPT Plus only recently introduced evolving and inpainting for Dall-E 3, and I think OpenAI is going to keep this exclusive for ChatGPT Plus subscribers for as long as it is profitable for them. 😑

The new (cheaper) SDXL turbo text2image models used simply don't support start images because of the way they have been engineered. I guess NightCafé opted not to implement image2image, because the performance/quality trade-off isn't quite worth it (yet). Who knows … there are lots of new fine-tuned models popping up, so it's likely that one featuring image2image will eventually make it into the NC catalogue.

They are good gatekeepers for the vast jungle of generating AI models after all. 🤷‍♂️

Thank you both for the information! I don't know the background. @knusperfunk @John_Shadeslayer

👍👍

Sounds interesting. I'll try it out. I'll also look at your gallery. 🙂

Yep. However now that it's been released, you can use SD3 along with Flux 1.0 for free something like 10x every 12 hours on Stablecog, then upload. It only takes a few seconds to sign up with a Google profile or email. They give you 100 free credits for joining. But since you read that, I should revise my glowing review. It's out now. Did it match Dall-E and make XL look like a dinosaur? Not really. While the picture quality is really nice, and it does adhere to prompts better than XL, the finished product took a huge leap backwards from the Beta I tested! Prompt adherence and creativity doesn't come anywhere close to Dall or Midjourney. I re-ran that tiger prompt to see how much better the finished version was. I was shocked and disappointed with what I got. It didn't give me everything I asked for. What it did give was typical SD fails with longer complex prompts. A blended macaw and toucan, and ridiculous fan shaped sprays of water coming from his mouth.

New

Beta

🤣

😁 many have, just not showing up yet

Theres a bug here, at the moment, sorry!

Oh, OK. I have 3 of them published now so far.

Should be fixed now! 🫣

Oh, OK. Thanks for letting me know. I can't wait to see other people's creations with this new model here when the bug is fixed.

OK, Thanks.

That's ridiculous. Another example of why nobody should waste credits on SC, and the fact that its Pro. Why? There is zero improvement, and no control. And it gives the same ugly, dull colors and "plastic" look as DreamSharper XL Light. NC has much better and cheaper models today.

Thank you for commenting. I actually was impressed with the healed scars. I was not a fan of the crown of flowers, but I guess He might prefer flowers after He was healed rather than thorns? I also would have preferred that, for this amount of credits, the flower border made sense to me. I received flowers on Him and a partial arbor - with good composition and balance. I agree wholeheartedly with you about the ugly dull colors that to me appear draft quality or somewhat textured. Why does He look more European than Jewish is another thought. It did provide a sturdy male that could have been a carpenter. Overall, I was somewhat disappointed with this image. I tried Stable Core a couple other times for other stuff back then if I remember correctly and deleted them. Not a big fan of the look this provides which to me looks like SDXL Lightning. No obvious errors on this like Lightning, but not a fan of the interpretation this provided. Some people might like it, but I prefer other models.

I guess it's a different model with different rules. I do agree that creating complex prompts for SDXL is better than that of the same prompt here. We probably don't have the credits to discern how best to use Stable Core (or if it's even that good).

It really cherry-picked what it could do best and ignored the rest of my prompt.