It's been like 3 months, and It's time... time for a NEW VERSION of the RAPID Workflow 🥳!!!!

I have brand new witchcraft for you.

Let me fill you in on the big stuff:

1. 10x speed increase with optimized cache routing for lightning quick batch image generation.

2. The workflow notes now have animations, buttons, links, and also there is a menu inside the workflow.

3. The custom prompt syntax is new, improved, and now works in general. There are no edge cases. I basically built a prompt compiler that compiles your prompt syntax using a custom prompt programming language (I seriously might have to implement some abstract syntax tree implementation of the prompt compiler if it gets any more complicated than it is now).

4. Everything can be toggled.

5. No more "!!UNDEFINED.safetensors" file is needed! You can just straight up run it now.

6. All the features from the previous version are still here, the implementation of each feature has been HEAVILY optimized. I mean, HEAVILY HEAVILY. They are TINY in comparison to how much processing they took up before.

7. ControlNets ARE HERE!! Get your controlnet workflow RIGHT HERE! BTW, I'm using this controlnet (https://huggingface.co/xinsir/controlnet-scribble-sdxl-1.0) for the previews that you see.

Why no 0.5.x version?

You might be wondering why there is no 0.5.x version. Well, about that. The long and the short of it is that there were so many things that I changed that I had to skip the 0.5.x series of updates. Now, for context, the last version of the 0.5.x updates was version 0.5.104. THERE WERE 104 VERSIONS OF 0.5! So, why did I change to 0.6? Because I discovered something that required an update so massive that I had to change versions. For perspective, the node count went from over 2000 nodes to less than 900 in the FIRST VERSION of the 0.6.x series of updates.

Why take 3 months?

Next question probably is, "Why did this update take so long?" Well, for as much as I wanted to keep adding features to the workflow, it desperately needed optimization. And guess what, that's mainly what took so long. I wanted to find the absolute limit of optimization, but no previous examples of such a feat existed. So, I had to do it myself. I looked through endless amounts of libraries, culled countless nodes, thought of countless optimization strategies, did way too many experiments, all to pretty much get somewhere I could have reached in a week if I had known what I do now. Other than that, I spent a whole month trying to come up with ideas for previews and came up with literally all of them within a day. I really do not know how to feel about this.

Final Notes:

Alright! That is pretty much all I have to say. Sorry this update took so long! If you want a quote on "Why you should choose this workflow?" here it is:

"The big deal is not that there is controlnet capability in this workflow, or that there is highres or prompt regions or custom prompt syntax or SD and SDXL toggling or notes or a menu, etc. It's that there is LITERALLY EVERYTHING ALL AT ONCE! It's all IN ONE PLACE!! AND it works at the fastest speed possible."

Thank you for using the workflow! Please give feedback! I'll make another update pretty soon depending on how much feedback I get. I'll also try to make posts myself and link the workflow as a tool. I forgot that was even a thing I could do up until now. Look forward to seeing some of that too I guess 👍.

Pre-Version v0.6.x:

Hello! I'm finally sharing the workflow I have been working on for the past 5 months now! It can do a fair bit of things and has a fair bit of features:

1. Organized workflow with labeled groups and short explanations using proper HTML notes. (They will look much nicer in the future, I just don't have the energy at the moment.)

2. Region specific prompts. (Inspired by the "Multiple characters" workflow.)

3. Randomized prompts.

4. Custom prompting syntax that allows commenting among other features.

5. Automatically adds the LoRA, the checkpoint, and all other prompt generation info to the final image on save.

6. Ability to switch between SD mode and SDXL mode. (Might work for versions >=0.4.6 idk)

7. Adds a way for denoise to work on SDXL models.

8. Outputs to a custom filepath controlled by what you put in the "Character" and "Version" nodes.

9. Ability to inpaint images.

10. Ability to use an image as latent.

11. Has HighRes that can be enabled or disabled at will.

12. (NEW) Adetailing pipeline as of version 0.4.6!

That's about all I can list for right now. I plan on adding many more features in the future.

!IMPORTANT: Create an empty file named "!!UNDEFINED.safetensors" and add it to your LoRA folder. The workflow uses it to check for empty or undefined LoRA's so it's very important that you do this!