Firstly, I’m by no means an expert not especially among the four-digits elites here on BW. I’m merely sharing my experience and creating a place for discussion among the lovely community here. (Feel free to use the comment space here for discussion)

Secondly, thanks to @terror2 and @heihund on the time you spent discussing about this model with me. This post was possible because of our conversations. (And thanks @terror2 for letting me use his format too)

Thirdly, most of these are covered in the instructions in the model but I’ve decided to make it a little visual here to demonstrate its potential and things to look out for.

Lastly, the prompts for this post are all visible to serve as examples and for observations. Images folded multiple times will have weird prompts.

===

FORGING

===

This is almost no different to how you do SDXL and Anima models except that you no longer have any lever settings on Sampler, CFG, and Steps (Which I used a lot for my prior works). In this environment, it is ChatGPT/Grok-like input which means natural sentences and language works best here.

You will probably end up writing an essay for your specificity so you might want to try FOLDING as explained in subsequent section

Because of the nature of this, the style is all over the place. Take a look at @terror2, @heihund, and @ranmafan experiments. There’s no style consistency and you will need to describe it yourself. This was a problem for me because personally, even though I am an anime fan, I have not watched any (series) for the past 16 years so there’s many artist or styles out there I don’t know who or how to reference. Some of you might already know how to control this. (I actually discovered Power, Makima, Miku, and Marin through BW and I made Victoria before I knew Marin so was a bit gutted they look alike)

The photo that Hattie showed for reference is here: https://betterwaifu.com/post/894o0im

===

FOLDING

===

In my opinion this is the real magic of the tool. You can use ingredients you have or make new ones with the cheaper SDXL and Anima models and fold them together using this tool. The possibilities of how you fold is up to your imaginations. You can use different image as anchor points and they output different results

Multiple layers of folding over the same string of output sequentially can result in some weird interpretation since noise is reintroduced every time you fold. #1 to #4 and #6 is an example of how the final image can drift with multiple sequential foldings

But sometimes it is necessary to do that in order to create a robust structured building block towards the final goal you had in mind.

With 389 credits, I think the yield is pretty high with SeeDream. Parallel Lines clocked closer to 800 credits with purely using Anima and tag bleeding was unavoidable. Also what I like about this tool is that I can start thinking of strategies how to fold things together rather than constantly generating in hopes that the AI knows what I want (Maybe I just need to prompt better lol). Over the next few weeks, I will try to demonstrate what SeeDream could do that I struggled earlier this year and there might be some reposts as well.

===

Thanks again and hope you all can see your dreams with this tool!

(My project got delayed so I also decided to delay Lexi’s debut for this. Sorry Lexi!)