What is the path of progress.?
They asked me in the chat:
Questions - why don't I use Lora to generate images, because this Coldflesh Realistic LCM model is very limited.?
Questions - Why is it that if the model is designed exclusively to generate women, I use it to generate the opposite, without the inherent possibilities.?
Questions - How do I manage to get something that was never expected from this model, which is my secret.?
It's not convenient to write this in a chat, so I decided to respond here in more detail.
Answer: The fact is that most people with an average level of training, and almost all people with an entry level of training, jump over difficult points in generation, no matter what it concerns - from building a workflow with thoughtful spaghetti management to writing a product. People want a quick result and that this result is easy for them. What does this approach bring in the end?
The result of this approach is at least a lack of the necessary experience for further progress, and at most such an approach will never bring professionalism.
What to do.?
The answer is quite simple.
Before switching from SD1.5 to other newer versions of SD, try to fully squeeze out of the checkpoint everything that is embedded in it on SD 1.5. Before connecting LORA, try to get everything you want from the Сheckpoint using Prompts.
Why this particular approach.?
Only this approach allows you to immerse yourself in all the subtleties and nuances.
Naturally, someone may object that there is no negative prompts in FLUX, so what is the reason for all this experience with prompts?
But in real life, if you are 15 yo, you won't be able to jump over 10 years [out of your 15 years] and immediately become 25 years old.
Maybe this example is too simple, but it reflects the very essence of progress: the gradualness and thoroughness of studying without the desire to jump over complex things and quickly get some result, even without understanding how this result turned out.
And these are not simple words - I will demonstrate this in practice using the example of working with a model Сheckpoint "Coldflesh Realistic LCM".
Questions: Why did you choose this particular model Сheckpoint "Coldflesh Realistic LCM" for your experiments.?
Answer: There are several reasons, and they are all obvious:
1. A small number of steps 8-12 is a unique case for SD 1.5.
You do not just save your time, you expand your capabilities and potential, because in other models, you will have to stop at 30-40 steps, and in this model, after just 6 more steps everything is just beginning.
2. The file size of the Сheckpoint model is only 1 GB.! While the entire size of similar Сheckpoint models is 6 GB.
3. Ample opportunities to use PROMT to obtain unique results, in comparison with other models.
I will not describe what the model Сheckpoint consists of, and how the process of extracting material from the Сheckpoint itself works - this is already well described on the Internet.
But it's worth noting the fact that the author of this model has surpassed all possible other models, given the file size and the number of steps to get results. For this, I thank him very much, and yes, thank him very much.!
When I finish squeezing all possible results out of this Сheckpoint without using LORA, you will be pleasantly surprised that what can be achieved, but when after all this I start connecting LORA to this model, your surprise will turn into delight.
You don't believe me.?
Then just follow my updates on image generation using this model "Coldflesh Realistic LCM".
Questions : Which is my secret.?
Answer: First, you need to figure out what the model consists of.
Next, you need to find out which tokens it uses.
After that, you need to figure out the power of the keywords. In the end, you have to reach the limit of the model's Сheckpoint capabilities, reach the limit when you are already running into a blank wall and cannot create anything that is required.
That's where the real work begins.
No need to abandon a Сheckpoint model and take on another one. Just jumping from one to the other will not gain you experience and knowledge.
You'll just get information, but information and knowledge are two different things.
At this very moment, when you have reached a seemingly impenetrable wall, you do not abandon the Сheckpoint model, and look for workarounds.
And you will find them.!
Because SD is a flexible system, knowing the principles of its operation, how it work, you can extract what is not initially embedded in the Сheckpoint model.
[P.S. If it takes you more than 10 minutes to generate one image, then you are moving in the wrong direction, you will not be able to progress and increase your level at this rate of generation. We need to look for reasons and completely change our approach.]
Thank you for reading.