Wait, a rundown? You mean a report, or?

@spookyghosts Much appreciated.

@makiaeveli I just wanna know the details of what makes it good, what makes it bad, @spookyghosts gave me a pretty good idea of it so far. Enough to make me interested to DL and try it out, atleast. 

@badhandproductions  lmfao it was a reference from the US "The Office" when Idris Alba asks Jim for a "rundown" and Jim's brain seems to break for no reason and not understand what a "rundown" could be. I mean spooky gave you a good answer, I was just joking Im sorry lol

You should also definitely try it out. I'm paraphrasing a random comment I read the the other day that said "this model could last 500 years". It does what you ask. If it could have finetunes, or if it's technology can be leveraged to produce a model that is fully NSFW, it is the best open model to date, period. F2K-9B can produce perfect skin, but only when you get magical parts of the prompt correct. This model is far more like ZIT or Qwen or even SDXL in its inherent understanding of the context of the scene. It's the best of all worlds such far ... minus edit. But you can use either F2K or Qwen to do that.

@makiaeveli thanks! I like the way you explained that. Could I ask you to elaborate on what specifically "edit" is for? Is it some kind of a more modern take on inpainting/outpainting? I remember it used to be a thing to have a setup where you paint a mask to get rid of something bad so you can prompt for and generate a bunch of replacements for the masked area. I guess what I'm saying is I'm not really comprehending how something could be more powerful than this, but ideogram def has my attention already by making bbox prompting the way to prompt.

@unphased nowadays some models are built on an architecture specifically enabling editing of images. It's not just simply deleting pixels with a mask and asking it to refill, the models are trained to know what is inside the image and alter it. Klein and Qwen have edit features. Of course, as you say, many models can simply take an input latent and continue sampling. But "edit" models are capable of understanding what you're asking for without bounding boxes or masks.

The difference is Ideogram takes all that masking information to begin with, essentially, to generate the image. It'd be like putting down your photoshop layers with text descriptions and having it fill them in. You aren't tricking the sampling process with multiple stages and masks, it does it all in one step. Currently, the way edit workflows work, you generate an image and slowly tailor it to your liking over iterations. With Ideogram, you should be able to get the exact output immediately.

@makiaeveli  Unfortunately I'm a philistine when it comes to kino TV, lmao