I have quite the coming together first set of prompt formula and tools I’ve come up with… still working on testing,. haven’t got to everyone of them yet most of them are repurposed SD1.5 tools we had just finished creating before SD1m5 went down… some kinks to work out in summary Flux01 uses a single 500 token generation Prompt read. while SD1.5 uses up to 10 or so maybe more or less depends not on tokens but overall computational resources allocated per user request. any way SD1.5 is harder to prompt too but is technically more versatile and with multiple prompt generation prompting SD1.5 gets about 600 to 800 tokens after accounting for the tokens you use up ect to bridge prompts… Flux01s issues are going to come down to keeping things separate, flux is mainly a prebaked blend on many levels and simplifications of generation, an upside is it won’t be nearly as hard as separating things in SD1.5, be cause Flux technically blends everything you’d think this be an even bigger issue, well flux is designed to better maintain specifications context contextual property and object identification prebaked into it with the Integer logical blending of the entire generation GAicomplex. I found that using containers certain ways can help that or separating mixes by using step loops. they both use the same syntax [] which can be subtly binded to things in parentheses out side it. usually out summary details in containers then hook them to the more messy stuff. []::;() that’s the syntax. “::;”… you could also get a separation object embodiment but strongly bind them as object details on Intial object body by using containers in series []::;[]

step loops work a bit different and you wanna look into the generation step counter of Flux01 to use them to their max potential, the easiest way to plot step loops would be my 2 to 3 / 321 rule. boxes with out step loops inside them are fundamentally a form of interpolationwith the system on more then just graphics level but the handling of the interpretation by the perchance LLM itself takes more of a play in handling some diner details of the “to system thrown box” [] so what’s my 321 rule¿ pretty simple if you can define a single object which has other things mixing with it you want to even out level the blending of for a proper blending of elements one easy way to do this is sort out all your details into things which have either 1 2 or 3 characteristics which aee of a layering mixing or blending or fusing criteria of, now then 1.you take the main ideas 💡 put them in an interpolation non step loops box [] take all your 2 swaps put them in a 2 step looped box one side for one half the other side for the other half / of all your criteria of the former which primarily have one opposing trade off from them selfs “2 swaps” [swap1|swap2] now take your criteria of things you could best squeeze into 3 groups of simmilar swaps or criteria ect out them in 3 swap… [||] the reason btw that this works so well is because inherently after each pass of either or loops the other one is exactly 1 out of alignment with the other in ratio typically 2 swaps should be your more profound swaps 3 sales are the misc tail end blends. this ensures that every 2 swaps is evenly blended with every 3 swap because 2 swaps and 3 swaps just work that way… some generator have 60 to 70 step counts to plot step loops properly though becomes a big issues when you are handling 5 to 19 or more logical swaps. hope you enjoyed the only good 👍 tip I got for Flux01 at the moment ciao 👋🌊