Software Engineering Courses (SECourses)•15mo ago

https://x.com/ostrisai/status/1842388844970135932 hes still working on it

https://x.com/ostrisai/status/1842388844970135932

hes still working on it

I haven't spoken much about my ongoing project to de-distill schnell to make a permissive licensed version of flux, but I have been updating it periodically as it trains. I just noticed it is the #2 trending text-to-image model on Hugging Face. Working on aesthetic tuning now.

Twitter

•

10/5/24, 2:18 AM

crystalwizard•10/5/24, 5:51 AM

it's not a matter of reverse engineering it, it's a matter of it's being impossible to undo

MMaxivi https://x.com/ostrisai/status/1842388844970135932 hes still working on it

crystalwizard•10/5/24, 5:51 AM

yeah, i know. he can't do what he think he can.

crystalwizard•10/5/24, 5:52 AM

it's sort of like this - if you mix up flour, water, eggs, chocolate chips, sugar, baking soda, salt - you have a batter that can be turned into a lot of things. YOu could make a chocolate chip cake or cookies at this point. if you then bake cookies, you have cookies. and no one can come along and turn those cookies back into batter.

crystalwizard•10/5/24, 5:53 AM

they can soak them in water, and get mush

crystalwizard•10/5/24, 5:53 AM

but it's still nto batter

crystalwizard•10/5/24, 5:54 AM

you can then shape the mush into some other shape, and say you turned it back into batter and now you've baked a cake or something - but you didn't actually do that. you just made mush and reshaped cookies into cake shaped cookies that now don't taste all that good

dsienra•10/5/24, 5:55 AM

if you right will see all this projects fail

MaxiviOP•10/5/24, 5:56 AM

i mean i think he's right, i tried 100+ trainings on these distill models, it just gives it texture and artifacts

Ddsienra if you right will see all this projects fail

crystalwizard•10/5/24, 5:56 AM

eh, he'll keep working on it till he gets something that he'll then prance around and boast about. you'd think HE would know better - but he's backed himself into a corner

crystalwizard•10/5/24, 5:56 AM

he's desperate to be able to create controlnet for flux - and you can't do that, either.

MaxiviOP•10/5/24, 5:57 AM

these dont work??
https://huggingface.co/Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro

Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro · Hugging Face

MMaxivi these dont work?? https://huggingface.co/Shakker-Labs/FLUX.1-dev-ControlNet-Unio...

crystalwizard•10/5/24, 5:58 AM

that depends on what you mean by work, and whether what ostris wants is what they actualy do or not

dsienra•10/5/24, 5:58 AM

let's see as I said I'm not an expert, I can't gave a informed opinion

dsienra•10/5/24, 6:00 AM

you have good points

Ddsienra you have good points

crystalwizard•10/5/24, 6:01 AM

i've got close to 3500 hours in on studying stable diffusion in the last 2.5 years. and close to 300 hours in on studying flux since it released. if not more

crystalwizard•10/5/24, 6:01 AM

i literally spend an entire month sitting here, feeding it one single prompt at a time to see what it would return by default - walking through it's latent space

crystalwizard•10/5/24, 6:02 AM

give it an unclear concept - you're going to get women, dogs, fantasy, or anime cat girls - and frequently a mixture of those

crystalwizard•10/5/24, 6:02 AM

not random, all over the place, images - like you should

crystalwizard•10/5/24, 6:03 AM

it's massively overfit, and it's been heavily doctored to deal with the 'woman laying in the grass' warping and shortening issue it shares with sd3-2b-medium.

dsienra•10/5/24, 6:03 AM

yes that is because of the distilled guidance right?

crystalwizard•10/5/24, 6:03 AM

and that makes it impossible to do anything with other than what it is

Ddsienra yes that is because of the distilled guidance right?

crystalwizard•10/5/24, 6:03 AM

no. that just makes it worse.

crystalwizard•10/5/24, 6:03 AM

no, that's becuase of other things that were done before that

dsienra•10/5/24, 6:03 AM

you always get something coherent

crystalwizard•10/5/24, 6:05 AM

sure, because it uses the t5xxl encoder nd the clip_l encoder. so you're going to get coherant. but give it a word like Umber and you should get mostly the color. the dictionary defintion for umber is "a natural pigment resembling but darker than ocher, normally dark yellowish-brown in color ( raw umber ) or dark brown when roasted ( burnt umber ).
2.
a brownish-gray moth with coloring that resembles tree bark."

crystalwizard•10/5/24, 6:05 AM

that's not what you get

crystalwizard•10/5/24, 6:05 AM

you should get random subjects, since umber isn't a subject - all over the map

crystalwizard•10/5/24, 6:05 AM

you don't get that

crystalwizard•10/5/24, 6:06 AM

you get very specific subjects as that's what it draws when you don't tell it a subject

dsienra•10/5/24, 6:06 AM

you put something random and and get a dog for example

dsienra•10/5/24, 6:07 AM

I know what you said

crystalwizard•10/5/24, 6:07 AM

if i tell you this word "Unboundedness" - what would you tell me it looks like?

dsienra•10/5/24, 6:07 AM

hehe

dsienra•10/5/24, 6:07 AM

no clue

dsienra•10/5/24, 6:08 AM

the model probably returns a cute anime cat

dsienra•10/5/24, 6:08 AM

hehe

crystalwizard•10/5/24, 6:09 AM

it should return random images with random subjects. because it doesn't have a subject so it should just be pulling up data from all over latent space

crystalwizard•10/5/24, 6:10 AM

if you put that word into sd 1.5, or sdxl, or midjourney - you'll get interesting results

crystalwizard•10/5/24, 6:10 AM

put it in flux dev and see what you get

dsienra•10/5/24, 6:11 AM

thanks for your detailed explanation, let's see how this ends up

Ddsienra thanks for your detailed explanation, let's see how this ends up

crystalwizard•10/5/24, 6:11 AM

based on the blowup that he had a few weeks back, he'll refuse to admit he cna't do what he wants. he'll continue on till he gets 'something' and then boast about it

crystalwizard•10/5/24, 6:12 AM

and it'll return some sorts of results, and people will believe he undistilled flux because they don't know any better. and blackforest will sit there and laugh at him

dsienra•10/5/24, 6:14 AM

I know that is imposible to convert a distilled model to its original state, the information that was lost in the process is lost forever

crystalwizard•10/5/24, 6:15 AM

that's true too. you spend the money, you no longer have the money

crystalwizard•10/5/24, 6:16 AM

we get requests along those lines for Photoshop all the time "hi, i have this photo of this car, can you make it so i can read the license' - and the plate is just a black and white blur. 'um, no. this isn't camera raw footage, it's a jpg"

dsienra•10/5/24, 6:17 AM

what could be a success is to make a model that is more flexible and more trainable, will not be the black forest pro model, will be other model that is more trainable

dsienra•10/5/24, 6:18 AM

that if its possible

Ddsienra what could be a success is to make a model that is more flexible and more traina...

crystalwizard•10/5/24, 6:18 AM

to do that, robin will need to scrap flux and start over. and that was at least a million, possibly more, to train. that's a huge hit now that he's nto at stability.ai

crystalwizard•10/5/24, 6:18 AM

so he's not likely to do that

crystalwizard•10/5/24, 6:19 AM

what he's most likely to do is continue to develop Pro - it's api only, it's making him money, he can sell it to top movie studios and commercial businesses

https://x.com/ostrisai/status/1842388844970135932 hes still working on it

Similar Threads