Everything explained here is for the purpose of organizational assistance. Notion is not a program that will connect to Stable diffusion. It is simply a way to create a document that will help you quickly control the LORAs you create and your activation words. You could do the same thing in Word or Excel.
To understand the different phases and everything else, it is crucial that you access the protocol created to inform you about how to create LORAs faster. Furthermore, it is recommended that you already have some basic knowledge of how to create a LORA
How to train MORE LORA models in LESS time and in a simpler way?
How do I duplicate this table and save it so that I can edit it myself in Notion?
To duplicate this table and manage your LORA projects more easily, you simply need to have an account on Notion, the same website where you are reading these notes. Notion is used to take notes and share them with others easily. Here is a small tutorial on how to duplicate the previous notes to keep them for yourself:
<aside> 👉 Step by step:
At the top bar, you should see an option called "Duplicate". If you don't see it, click on the three dots.
it redirects you to the desktop or phone app or to your workspace from the web itself.
Before showing you the tables, I'll give a brief explanation of the sections and their possible uses in the different phases of LORA creation.
This will be the table we use the most. It is used to quickly switch between different phases of LORA creation. Its sections include "Add new LORAs", "Phases", "Phase 1", "Phase 2", "Phase 3", "Phase 4", "Failed management", "Finishing", and "General list".
Add new LORAs: This is a quick way to add new LORA projects. You can use it during Phase 0: preparation. You can add images in the "image" section, select priority, and add tags to filter projects. do this, simply click on the empty boxes.
<aside> ☝ To get to the next table, we need to change the view. We can simply click on the top part where the current table name is.
</aside>
Phases: You will switch here to move LORAs quickly between different phases as you complete them (you just have to drag them). If you do the trainings collectively (i.e., if you do several datasets/taggings of LORAs in a row), will be easy to do.
<aside> ☝ But if you are going to train only one LORA and constantly move it between phases, it will be annoying to keep constantly changing. In that case, I recommend directly adding a section in each table of each phase to move LORAs quickly. To do this, select the three dots, go to properties, and in the "Hidden in table" section, click the eye icon of the "Phase in which it is" property and there you go! You should now be able to change phases in the table where you enabled it.
</aside>
Phase 1: Here, you can quickly select whether the dataset is manual (you looked for and curated it yourself) or if it was created via Colab. In the "Annotations" section, you can include information such as which tags were excluded in the search, and notes about the dataset quality. You can also track the total number of images.
Phase 2: Here you can quickly classify a LORA whether it is tagged with a realistic model (Blip) or an anime type model (Waifu diffusion tagger). In these annotations, you can include information such as how many tags were removed and merged with the trigger word. Plus, I added a section where you can add it and quickly access it in Phase 4.
Phase 3: In this phase, we can classify the type of LORA we made and quickly check how many we have in the dataset to calculate the number of epochs. In annotations, we can leave any parameters we want to highlight in the training
Phase 4: In this table, we can quickly copy the Trigger word for testing. In annotations, we can add the epoch we used or other notes such as what exactly went wrong in the tests. Finally, once we have done the 10 tests, we can classify the result based on whether it had errors. If it had a few minor errors, we can leave it in acceptable, but if it had many or very serious errors, we can put it as "Repeat" or "Try with other epoch". I have left another table below for tests with more than one epoch.
Failed Management: Here go all the LORAs that were marked as "Repeat" in the previous phase. You can quickly see the annotations from each phase in case they were useful to see where you may have failed. You also edit them for the next training. Finally, do not forget to change the phase back to where you concluded that you failed. An example would be my Kamina's LORA in beta version, where I concluded that I failed in the tagging, so I should change it to "Phase 2".
