path.md

The path I took:

1. Setup the repo to work in it
2. Turning the model into a diffuser model
3. Turned it into a pipeline
4. Loaded the diffusor model

• Trying these models:
  • clarity_3
  • dreamshaper_8
  • realisticVision_6

5. Wrote argparser for CLI
6. Finish the path from the article
7. Need to fix the argparse in main.py
8. Need to add save location
9. Add real-ESRGAN
10. Error Handling
11. Lora

    Try these:

    • negative embeddings lora's
12. Make a model selector menu that reads models from the diffused folder
13. Rehashing the code:
    • Check if the pipeline setup inside a function is an issue, should not be an issue because the embdedding one works, will have to see about the global effect of using EADS in it or not
    • Debug the masks and the emb's 117-122
    • Remove the commented code, shift learnings to a ignored file
    • The changes that I made:
      • Took lines 167-170 from inside pipeCreate
      • added .input_ids
      • replaced getEmbdedding's main core logic, what a shame
14. Fix the output screen
    • Show the prompt, the neg prompt, the seed, cfg scale
    • Load the above details into the pictures (check civit.ai)
15. Show save location
    • Make a save folder logic, checks for exist
    • The name for the file saving
16. Try adding a progress loading bar
17. Check how to change the sampler
18. For the main task:
    • Need to make one model that gen's the BG and the other that gen's the FG
    • Or I can use a singular model for this
      • Use DeepLabv3+ for the mask
      • Pass that to the BG generation model
      • Assign the mask image to a variable
        • Save the mask, name is temp_mask inside output
      • Dump memory to clear space in the end off the VRAM after gen is done
      • Add inpainting model option to argparse
        • Add BG prompt to the CLI as well
      • Make this a function that is called
      • Have to push to main pipeline_setup file
    • Or I can gen the FG, cut it out using UNet, and then gen a BG
19. Add a method that upscales the image, realESRGAN
20. Add a method that fixes the faces, GFPGAN
    • Do the repo setup from the readme
    • Do a demo run
    • Add the models from the readme if needed
21. Try to add controlNet inpainting
    • Can add controlNet Aux for posture detection
22. Add a method that saves the image generated with a pre-fix
23. Add upscale to the CLI
    • Make a interactive UI, which shows size of image and then
24. Fix the size issue in the embeds
25. Fix issue of single seed being used
26. Make the readme.md with examples, code hashes, pictures
    • Give credits to the projects used, and the citations needed
    • make the commands better visible
    • Add link to install civit.ai Models, and the ones that I have used