Deliberate is the only Stable Diffusion model whose creator doesn’t just mix existing models but actually trains it. What you see on other sites are models by authors who merely merge different LoRAs and mix models together. To prove this, you can compare the changelog of Deliberate with those of other models. The creators of other models have to concoct something new for their blindly following fans every time, but you won’t see in any changelog that an author has added or removed something without changing the structure of the entire model. Everyone is essentially stealing from the creator of Deliberate, as within the entire Stable Diffusion community, no one else is training models, and there are no instructions for either training or finetuning available. My personal guide on creating your own models will never be publicly released as long as the majority continues to endorse mixes, pseudo Turbo models, pathetic attempts at XL, and fails to acknowledge whom they should be thankful to for even having the opportunity to mix their crap in the first place. You might think this text is filled with toxicity, but you have no idea how much of my personal time and money has gone into researching, buying the most professional equipment to create datasets for training Deliberate. And then, everyone just mixes it with a couple of clicks and collects donations for it. If you’re interested in supporting the development of Deliberate and want it to remain free and truly competitive with DALL·E and Midjourney, please donate using this link.
You can now forget about extremely detailed, 8k, hyperdetailed, masterpiece, etc. You can still use them, but it’s not necessary, because you’ll get a masterpiece anyway. Just take a look at the demo images and their EXIF data.
Masterpiece tip:
If you still gonna use masterpiece token in your Prompt, trust me, the real power of this token you can see only if you keep your Negative prompt empty.
By using a ton of tokens in the Negative, you’re seriously limiting the model’s ability to create a variety of images and faces. Plus, using someone else’s Embeddings with built-in tokens is a really bad idea. The creators of public Embeddings never spell out what they’ve put into them, and you have no idea what tokens are in there. They could have a word whose presence in the Negative would stop you from getting the art you want. Like morbid or deform. It’s way smarter to use the ones that are right for each specific case. The homepage has loads of examples of how to use Negative the right way.
You can boost the result by adding the mj token (at the start of the prompt), cozy or cinematic.
You probably guess what mj means and how the magic works. So you have it.
I’ve completely revamped the cinematic token. As you can see, Deliberate v6 always strives to find reflections and create stunning results (without crazy prompting), much like Midjourney does. At the same time, sometimes we want to achieve the highest level that no photographer can reach. That’s why I’ve redefined the interpretation of cinematic.
Oh yeah, I’ve rebuilt many of the NSFW tokens to deliver perfect results with just one word. You can find all the details on my Boosty page.
The dataset is truly enormous. In fact, this is the first public model on the internet, where the selection of images was stricter than anywhere else, including Midjourney. Deliberate v6 can work without negatives and still produce masterpieces. This became possible precisely because of the huge dataset. It’s unique, it’s massive, and it includes only perfect images. Just try typing any word, exclude the negatives, and you’ll see that Deliberate knows what to show you without randomness. Even when you’re trying to get some ugly result, it still provides you with a spectacular one.
The license type is CC BY-NC-ND 4.0. If you want to use Deliberate on your service and/or for commercial purposes, message me on Discord. If you are an individual, not a commercial organization, there are no restrictions for you. The philosophy is very simple – private individuals provide support for the model, not companies, so private individuals can do anything. When using the model and any derived materials, it is mandatory to credit the author.
Now you can stop arguing with people to prove what the best sampler in the world is. You can prompt my nickname XpucT on a 512 x 512 and see how good the image looks, count how many fingers the girl has, and check if there is correct sign text or not. From now on, you can just run batch Euler a and UniPC, for example, and see who make the job better.
All the prompts on the homepage are my own and serve not just to show off what the model can do, but also to help you get whatever you want, using just one Deliberate model. Unlike the pics you see on Lexica or CivitAi, my prompts always hit the mark. They’re not just a jumble of nonsense words, no random lucky generations like what you find on those sites.
