When AI sees a mirage, an object, and image many enough times, it starts to dream it more vividly. Here, my best friend is sent to make peace to the world. Introducing Madina to the Stable Diffusion. The model was trained in 30 minutes by Google Colab and the tool to generate pictures is called InvokeAI. With InvokeAI, the base model used is Stable Diffusion 1.5. The first picture gives the reference of Madina. It's seamingly difficult to prevent doubles or even triples.
InvokeAI
Stable Diffusion 1.5

Continuing with the theme of own trained pictures. This time testing different art style, Hokusai.

Madina as a cyborg in a dark science fiction city, detailed face.

Here we are the God, having the power of voice to generate almost any kind of images. Feels very strong without any filter with the Stable Diffusion, but we should also carry some responsibility. However, the art generated can even be very mind-blowing. If you want horror, AI is an expert. Right from your most vivid nightmare during the highest fever. But hidden humane feelings that words or human brush sometimes cannot catch are sometimes a result of a Stable Diffusion artist.

You can express your ideas quite freely with Stable Diffusion, there seems to be mysterious touch but still the message comes through. "Praise people that are different, and stop discriminating aliens."

Training Stable diffusion with pictures bring possibility to bring your old, passed away, but loved pets into new surroundings, even play with you in the pictures. I don't recommend playing too much, as it can be heart-breaking, but yet another approach to use AI-generated pictures (and later videos) to bring back some people and animals, long passed.
The first picture displays my original friend.

How well can monuments and landscaped be taught to stable diffusion? There can be a lot of tweaking and prompt refining needed to get better results than below. However, the tower here (original in the first photo) is somewhat recognizable. One image is an actual watercolor painting of the tower.

Here we display variations of Sapokan vesipuisto. The first picture shows the original waterfall. Last one is an impression with watercolors.

Some technical details of painting with Stable Diffusion. Stable Diffusion 1.5 has a wide range of so called LoRa (low-rank adaption) models which fine-tune existing models. No need to train a full model, like in Dreambooth. Here is one example of using LoRa model "(r1ge - AnimeRage)": https://civitai.com/models/115100/rage-unleashed. Samples may not introduce the best way differences, but you could add styles, characters or details with LoRa models that are of small size. First image without LoRa (rage), second one with rage effects and lightings. Prompt: "happy monk meditating in cyberpunk city, glowing, magic, aura, energy, style of Richard Anderson art."

Ready to rival with ShutterStock and iStock? AI text-to-image models, Stable Diffusion being one of them, pose a real threat to their business models, I'm afraid also to the photographers, models and artists. Soon it's difficult to differentiate AI-generated photos from real ones. So far, there are certain areas that still reveal Stable-Diffusion- 1.5-generated pictures but, with a bit of fine-tuning prompts, even SD 1.5 generates terribly authentic pictures. AI-generated pictures:

One hidden gem of AI-painting is possibility to paint inside existing image. So, for example, you could paint a happy group of xmas elves dancing into your yard or just paint a polar bear to sit next to grandpa. There are special models for inpainting, however, I just used my own trained model and standard Stable Diffusion models (1.5 + 2.1) to generate the following suggestions of murals to our town hall.

Inpainting makes it easier to combine several concepts. Althrough it's possible, there is tendency to duplicate the color themes and even disregard other concept if it's a rear combination in the model.