Still learning. There seems to be no easy way of learning this. The tutorials are all out of date or rudimentary or else narrated by someone in a droning monotone that puts me to sleep. So I learn by doing, and after many failed experiments, I have picked up on a few things.
The easy way is . . . did I say there was no easy way? I did, but this is the best I have so far. Start by just cruising the Tensor.art and looking at all the pretty pictures, models and loras, and press the star in the upper right hand corner of those that particularly please you or you might want to use in the future. There are hundreds and hundreds of these, so don’t pick all of them, only the ones that stand out.
To start producing AI press “Create”. If you aren’t yet registered at the site, there will be some hoops to jump through. On the next screen, in the upper left corner, there will be an illuminated red button, “SD3 & DiD”. Don’t touch it, it just means you are producing things from text; we will discuss the other options some other time.
Further down on the left side, near the middle of the page, you will see the word “Models”. This is where you add your first model, which will set the pattern for the rest. By now, you should have some idea of what you want to produce. Remember all those pictures I told you to star? When you press the arrow, you will be presented with a choice of models arranged by category. On the far left side of the list, you will see a button, “My Starred”. Click there, and you will see all those pictures you just starred, or at least the ones that are suitable as models.
Here things get tricky. You need to stick to a consistent style. Some are realistic, and others look like cartoons or anime, and some have names like “mix” or “fusion”. Notice that to the lower left of each picture is a designation like “v1.0” or “v2” or number up to 16. These are different versions that may not be compatible with each other, so it is good to know which one you are working with. Combining incompatible versions can produce an awful mess, and you will be seeing many of these before you master the technique. Pick whichever one pleases you, and press “Select”. Or, if none of them is what you want, you can go back and sift through the different categories until you find something.
In this case, I picked “LoRaLover_XL -V2” to be my model. The picture looks a bit like what I was trying to create, so I figured I had a head start, and would not have to struggle against my own model. XL means extra large, and I assumed that it would be easier to accommodate a small format in an extra large one, rather than the other way around. V2 is not the top nor the most popular format, which would be V1, but I thought it would be close enough that compatibility wouldn’t be a big problem.
Now that you have your base model, go down a bit and press the button that says "Add LoRA". This brings you back to the picture categories. Press “My Starred” and you will have a choice of things you can add. IMPORTANT: So long as you are in “My Starred”, you will be blocked if you try to choose an incompatible lora. But if you don’t find what you need there, you will have to choose another category, with no guarantee that anything there will be compatible.
In this case, I chose three loras: one each for clothing, position and background. I could have added a lot more, but the more things you add, the harder it will be to fit them to your composition and the greater the chance of incompatibilities. Above your loras, is a box, “Prompt” where you can right a short text description of what you want. You may notice there is already some text written there, provided by the loras themselves, and you shouldn’t change that in any way. Below “Prompt” is “Negative Prompt”. I rarely write anything there. Some people cram it with things like “ugly” or “low quality”, but I don’t think it makes much difference.
Once you have the loras you need, press the big green button that says “Generate”. Depending on your priority and how many people are using the server, it may take a long time or a very long time to get your AI image. You can speed things up by choosing “1” in the box next to the green button, which means that only one image will be generated, but your chances of getting what you want on the first attempt are low, so I usually choose “4”. (NOTE: If you have exceeded your quota of images, they won’t tell you that, they will just leave you to looking at a screen that says “Waiting” until you give up and go to bed.)
OK, now you have 4 images. You can simply click one, download it and you’re done. But wait! Why does she have three legs??? It is not unlikely that ALL your initial pictures will be flawed in some way, which means you will have to try again. The simplest kind of problem to fix is when some detail in a lora fails to show up or is indistinct. Your brilliant AI bot is only a machine, after all, and it had a lot of balls to juggle, and it didn’t give that one enough priority. Look at the listed loras. Each has a slider to the right, which you can adjust to give that lora more or less prominence in the next attempt.
Or you may find that two of your loras are incompatible for some reason. Take one out and start over. BUT be careful, if you keep changing things, the cumulative changes may pile up and turn into something really monstrous, and if that happens, the best solution is to delete everything and work on something else for awhile.
As you can see, in this case, I got three pretty good pictures and one worth uploading. That’s what I consider an optimum result, but it doesn’t happen every time.