The Free & Uncensored Version of MidJourney! (FLUX.1)

Matt Wolfe
6 Aug 202418:24

Summary

TLDRDieses Video präsentiert Flux One, ein neues AI-Bildgenerierungswerkzeug von Black Forest Labs, das von den Schöpfern von Stable Diffusion entwickelt wurde. Es gibt drei Modelle: Flux One Schnell (für lokale Entwicklung und persönliche Nutzung), Flux One Dev (für nichtkommerzielle Anwendungen) und Flux One Pro (für Enterprise-Lösungen). Flux One zeichnet sich durch Realismus, Textgenerierung und prompte Adhärenz aus, obwohl es in der Darstellung von Illustrationen und künstlerischen Stilen wie Öl- oder Wasserfarben noch Verbesserungsbedarf hat. Das Open-Source-Modell Flux One Schnell könnte zukünftig durch Community-Beiträge weiter verbessert werden.

Takeaways

  • 😀 Ein neues KI-Bildgenerierungswerkzeug namens Flux One ist von der Firma Black Forest Labs entwickelt worden.
  • 👨‍💻 Das Team hinter Flux One umfasst Experten, die an der Entwicklung von Stable Diffusion beteiligt waren, einschließlich der Erstellung von VQGAN und Latent Diffusion.
  • 🔢 Es gibt drei Modelle von Flux: Flux One Schnell (für lokale Entwicklung und persönliche Nutzung), Flux One Dev (mittleres Modell für nichtkommerzielle Anwendungen) und Flux One Pro (oberes Modell für Enterprise-Lösungen).
  • 🆓 Flux One Schnell ist unter der Apache 2.0-Lizenz verfügbar, was bedeutet, dass es quelloffen ist und kommerzielle und nichtkommerzielle Nutzung ermöglicht.
  • 🌐 Es gibt mehrere Websites, die Flux integriert haben, und einige erlauben es, die Modelle kostenlos zu nutzen, einschließlich der Hugging Face-Plattform.
  • 🎨 Flux One ist gut darin, realistische Bilder zu erzeugen und ist in der Textgenerierung überlegen, aber es scheint bei der Erstellung von Illustrationen oder Kunststilen wie Ölgemälden oder Aquarellen nicht so stark zu sein.
  • 📈 Flux One kann komplexe Anforderungen an die Bildgenerierung bewältigen, aber es kann bei der Prompt-Einhaltung und der Realismusdarstellung mit anderen KI-Modellen wie Mid Journey oder DALL-E 3 konkurrieren.
  • 🚫 Flux One hat zwar eine Filterung für nicht sicher für die Arbeit (NSFW)-Inhalte, ermöglicht aber die Erzeugung von Bildern mit urheberrechtlich geschützten oder lizenzierten Materialien.
  • 🔄 Die Open-Source-Natur von Flux One Schnell wird es anderen Entwicklern ermöglichen, das Modell zu verbessern und anzupassen, was zu einer Steigerung der Bildqualität und -genauigkeit führen kann.
  • 📹 Black Forest Labs hat angekündigt, dass Flux One auch als Grundlage für zukünftige text-zu-video-Modelle dienen wird, was KI-basierte Videoerstellung für Benutzer eröffnet.

Q & A

  • Was ist Flux und wer hat es entwickelt?

    -Flux ist ein neuer KI-Bildgenerierungswerkzeug, entwickelt von Black Forest Labs. Das Team hinter Flux umfasst viele Mitglieder, die an der Entwicklung von Stable Diffusion beteiligt waren, einschließlich der Erstellung von VQ-Gan und latent diffusion.

  • Wie viele Modelle gibt es in der Flux-Serie und wie unterscheiden sie sich?

    -Es gibt drei Modelle in der Flux-Serie: Flux One Schnell, Flux One Dev und Flux One Pro. Flux One Schnell ist das schnellste Modell für lokale Entwicklung und persönliche Nutzung, Flux One Dev ist effizienter und für nicht-kommerzielle Anwendungen vorgesehen, und Flux One Pro ist das leistungsstärkste Modell für Enterprise-Lösungen.

  • Unter welcher Lizenz steht das Flux One Schnell Modell und was bedeutet das für dessen Nutzung?

    -Das Flux One Schnell Modell steht unter der Apache 2.0-Lizenz und ist Open Source. Dies bedeutet, dass es kostenlos genutzt, verändert und verteilt werden kann, einschließlich kommerzieller Nutzung.

  • Wo kann man Flux-One kostenlos testen?

    -Man kann Flux-One kostenlos testen, indem man sich auf Hugging Face bei Black Forest Labs bewegt und dort die Schnell- oder Dev-Version im Rahmen von Hugging Face Spaces verwendet.

  • Was ist Glyph und wie kann man es verwenden, um mit Flux zu arbeiten?

    -Glyph ist eine Plattform, die als AI-Workflow-Builder dient. Es ermöglicht es Benutzern, ihre eigenen Flux-Workflows zu erstellen und sogar das Flux Pro Modell kostenlos zu nutzen, um Bilder zu generieren.

  • Wie gut ist Flux bei der Erzeugung realistischer Bilder im Vergleich zu anderen Modellen?

    -Flux ist gut bei der Erzeugung realistischer Bilder und wird als ähnlich gut wie Mid Journey bewertet, vor allem wenn die richtigen Prompts verwendet werden. Es ist besser als frühere Versionen von Stable Diffusion, aber noch nicht so gut wie Dolly 3 bei der Prompt-Einhaltung.

  • Was sind die Stärken von Flux bei der Textgenerierung?

    -Flux ist sehr gut bei der Textgenerierung und kann detailreiche und komplexe Texte in Bilder umwandeln, wie zum Beispiel Logos oder selbst Portraits mit Text. Dies ist ein Bereich, in dem es sich von anderen Modellen abhebt.

  • Ist Flux in der Lage, Copyright- oder urheberrechtlich geschützte Bilder zu erzeugen?

    -Flux ist derzeit nicht in der Lage, nicht sichere für die Arbeit (NSFW) Bilder zu erzeugen, aber es kann理论上 generate copyrighted images oder Bilder mit existierender IP, solange es keine spezifischen Einschränkungen für diese Art von Inhalten hat.

  • Was bedeuten die offenen Quellen von Flux One Schnell für die zukünftige Entwicklung?

    -Da Flux One Schnell Open Source ist, können andere Entwickler es herunterladen, anpassen und verbessern. Dies bedeutet, dass es zukünftig noch leistungsfähigere und vielseitigere Anwendungen für Bildgenerierung geben wird.

  • Welche weiteren Funktionen hat Black Forest Labs für Flux geplant?

    -Black Forest Labs hat angekündigt, dass Flux auch als Grundlage für zukünftige text-to-video-Modelle dienen wird, ähnlich wie die von LUM's Dream Machine und Runway gen 3 Sora, was bedeutet, dass es bald eine Open-Source-Option für solche Generativsysteme geben wird.

Outlines

00:00

😎 Einführung in Flux One von Black Forest Labs

Der erste Absatz stellt Flux One vor, ein neues AI-Bildgenerierungswerkzeug von Black Forest Labs. Es wird betont, dass das Tool von vielen Mitgliedern des Teams entwickelt wurde, das auch an Stable Diffusion beteiligt war. Flux One umfasst drei Modelle mit unterschiedlichen Leistungsstufen und Kosten: Flux One Schnell für lokale Entwicklung und persönliche Nutzung, Flux One Dev für nicht-kommerzielle Anwendungen und Flux One Pro für Enterprise-Lösungen. Flux One Schnell ist unter der Apache 2.0-Lizenz verfügbar, was bedeutet, dass es quelloffen ist und kommerzielle Nutzungen ermöglicht.

05:02

🖌️ Flux One im Vergleich zu anderen AI-Bildgenerierungen

Dieser Absatz vergleicht Flux One mit anderen AI-Bildgenerierungswerkzeugen. Es wird erwähnt, dass Flux One in Bezug auf Realismus gut abschneidet und in der Textgenerierung sogar überlegen zu sein scheint. Es wird auch auf die Fähigkeiten von Flux One hingewiesen, detailreiche und textbezogene Anforderungen in Bildgenerierungen zu erfüllen. Allerdings wird auch darauf hingewiesen, dass es in der Darstellung von Illustrationen und klassischen Gemäldestile wie Öl- oder Aquarellmalerei nicht so stark ist wie einige andere Modelle.

10:03

🚀 Flux One als quelloffene Option für KI-Bildgenerierung

In diesem Absatz wird die Open-Source-Natur von Flux One Schnell hervorgehoben und die potenziellen Vorteile für die Benutzer diskutiert. Es wird erwähnt, dass Flux One Schnell bereits in verschiedenen Websites integriert ist und kostenlos genutzt werden kann. Es wird auch auf die Möglichkeit hingewiesen, komplexere Workflows mithilfe von Glyph zu erstellen, wobei sogar das Flux One Pro-Modell kostenlos genutzt werden kann. Der Absatz schließt mit der Aussicht, dass Flux One in Zukunft zu einer text-to-video-Modell-Serie aufbauen wird, was es zu einer umfassenden Open-Source-Lösung für generative KI-Systeme machen könnte.

15:07

🌟 Zukunftsaussichten und Entwicklungspotenzial von Flux One

Der vierte Absatz befasst sich mit den zukünftigen Möglichkeiten und dem Entwicklungspotenzial von Flux One. Es wird betont, dass Flux One die Stärken anderer Modelle wie Mid-Journey, Dolly 3 und Stable Diffusion kombinieren könnte. Es wird auch auf die Tatsache verwiesen, dass Flux One bereits besser als einige bestehende Stable Diffusion-Modelle ist und sich in Kürze durch die Arbeit von Entwicklern weiter verbessern könnte. Der Absatz endet mit einer Aufforderung an die Zuschauer, das Video zu liken und zu teilen, um über neue Entwicklungen in der KI-Welt auf dem Laufenden zu bleiben.

Mindmap

Keywords

💡Flux One

Flux One ist ein neues AI-Bildgenerierungswerkzeug, das von Black Forest Labs entwickelt wurde. Es ist im Video als ein Werkzeug vorgestellt, das in der Lage ist, hochwertige Bilder zu generieren und in einigen Fällen sogar bessere Ergebnisse als Mid Journey zu liefern. Das Video untersucht die verschiedenen Modelle von Flux One, ihre Leistungsfähigkeit und ihre Anwendungsmöglichkeiten.

💡Black Forest Labs

Black Forest Labs ist das Unternehmen hinter der Entwicklung von Flux One. Das Team umfasst Experten, die an der Erstellung von stabilen Diffusionsmodellen für Bild- und AI-Video-Generierung mitgewirkt haben, wie zum Beispiel VQ-Gan und latent diffusion. Das Video betont die Erfahrung und das Know-how des Teams hinter diesem neuen AI-Tool.

💡Stable Diffusion

Stable Diffusion ist eine Gruppe von AI-Modellen, die für die Bildgenerierung verwendet werden. Im Video wird erwähnt, dass einige der Entwickler von Flux One auch an der Entwicklung von Stable Diffusion beteiligt waren, was auf die Qualität und das Potenzial von Flux One hindeutet.

💡Flux One Schnell

Flux One Schnell ist die schnellste und kostengünstigste Version des Flux One-Modells, das für lokale Entwicklung und persönliche Nutzung konzipiert ist. Es steht unter der Apache 2.0-Lizenz und ist quelloffen, was bedeutet, dass es für kommerzielle und nicht-kommerzielle Zwecke verwendet werden kann.

💡Flux One Dev

Flux One Dev ist eine mittlere Version des Flux One-Modells, die effizienter als die Schnell-Version und mit besserer Prompt-Adhärenz und Leistungsfähigkeit ist. Es kann für nicht-kommerzielle Anwendungen verwendet werden, aber es ist nicht erlaubt, Tools, die auf dieser Version basieren, zu verkaufen.

💡Flux One Pro

Flux One Pro ist die leistungsstärkste Version des Flux One-Modells und wurde für Enterprise-Lösungen konzipiert. Es bietet erstklassige Leistung und ist für die Nutzung in kommerziellen und fortgeschrittenen Anwendungen vorgesehen.

💡Glyph

Glyph ist eine Plattform, auf der Benutzer Flux-Modelle kostenlos verwenden können, einschließlich der Pro-Version. Im Video wird gezeigt, wie Glyph verwendet wird, um Workflows zu erstellen und Bilder mit Flux One-Pro zu generieren, was die Flexibilität und die Benutzerfreundlichkeit von Flux One betont.

💡Prompt Adhärenz

Prompt Adhärenz bezieht sich auf die Fähigkeit eines AI-Modells, alle Elemente eines eingegebenen Befehlstexts (Prompt) in der generierten Ausgabe korrekt wiederzugeben. Im Video wird Flux One mit anderen Modellen wie Mid Journey und Dolly 3 verglichen, um zu zeigen, wie gut es in der Prompt-Adhärenz abschneidet.

💡Realismus

Realismus ist ein wichtiger Faktor bei der Beurteilung von AI-Bildgenerierungswerkzeugen. Im Video wird Flux One für seine Fähigkeit, realistische Bilder zu erzeugen, gelobt, und es wird diskutiert, wie gut es in dieser Hinsicht zu anderen Modellen wie Mid Journey und Dolly 3 ist.

💡Open Source

Als Open Source ist ein Produkt oder eine Technologie gemeint, die frei verfügbar ist und von der Gemeinschaft weiterentwickelt werden kann. Flux One Schnell ist als Open Source lizenziert, was bedeutet, dass es von anderen Entwicklern genutzt, angepasst und verbreitet werden kann, was zu einer schnellen Verbesserung und Anpassung des Modells führen kann.

Highlights

Ein brandneues AI-Bildgenerierungswerkzeug namens Flux One ist auf den Markt gekommen.

Das Werkzeug soll sich mit Mid Journey messen können und in einigen Bereichen sogar überlegen sein.

Flux One wurde von vielen Mitgliedern des Teams entwickelt, das auch Stable Diffusion aufbaute.

Das Team hinter Flux One umfasst Experten für AI-Bild- und Videogenerierung.

Es gibt drei Modelle von Flux, wobei jedes leistungsfähiger und teurer ist als das letzte.

Flux One Schnell ist das schnellste Modell, offen für lokale Entwicklung und persönliche Nutzung.

Flux One Dev ist ein mittleres Modell, effizienter und prompt-adherenter als Schnell.

Flux One Pro ist das leistungsstärkste Modell, für Enterprise-Lösungen konzipiert.

Es gibt Websites, die Flux integriert haben, und es kann kostenlos verwendet werden.

Einfache Verwendung von Flux One Schnell und Dev über Hugging Face Spaces.

Glyph ist eine Plattform, auf der man Flux-Workflows erstellen und sogar das Pro-Modell kostenlos nutzen kann.

Flux One ist gut bei der Erzeugung realistischer Bilder und bei der Textgenerierung.

Flux One hat Schwierigkeiten mit der Erzeugung von Illustrationen, Ölgemälden und Aquarellen.

Flux One kann keine NSFW-Inhalte erzeugen, aber es ist nicht vorhersehbar, was in Zukunft möglich sein wird.

Flux One ist gut in der Prompt-Einhaltung und kann komplexe Anforderungen in einem Prompt bearbeiten.

Flux One hat eine Dual-Encoder-Technologie, die sowohl einfache als auch komplexe Prompts verarbeiten kann.

Flux One wird auch als Text-to-Video-Modell dienen und eine offene Option für generative Systeme bieten.

Flux One könnte die Stärken von Mid Journey, Dolly 3 und Stable Diffusion in einem Modell vereinen.

Die Zukunft von Flux One sieht vielversprechend aus, da es offensichtlich besser ist als vorhandene Stable Diffusion Modelle.

Flux One Schnell ist quelloffen, was bedeutet, dass es von anderen Entwicklern weiterentwickelt und verbessert werden kann.

Transcripts

play00:00

so there's a brand new AI image

play00:01

generating tool in town there's been a

play00:03

lot of claims out there that it is up to

play00:06

par with mid journey and in some cases

play00:08

even does a lot of things better than

play00:10

mid Journey so in this video I want to

play00:13

explore flux one from this brand new

play00:16

company called Black Forest Labs before

play00:18

I get into testing and experimenting

play00:20

with flux one here's a quick overview of

play00:22

some of the important details that you

play00:24

should know about this specific model

play00:27

the first thing that you should know is

play00:28

that it was actually developed by many

play00:30

of the team members who helped build

play00:32

stable diffusion among the team you can

play00:34

see our Innovations include creating VQ

play00:37

Gan and latent diffusion the stable

play00:39

diffusion models for image and AI video

play00:41

generation like stable diffusion XL

play00:43

stable video diffusion and rectified

play00:46

flow Transformers so the team behind

play00:48

this is a team of rockstars with a lot

play00:51

of experience in creating AI image

play00:53

generation and AI video models the other

play00:56

important piece of information that you

play00:57

should know is that there are three

play00:59

models on flux and each one is a little

play01:01

bit more powerful but also a little bit

play01:03

more expensive to use than the last so

play01:06

there is flux one Schnell which is the

play01:10

fastest model and it's designed for

play01:12

local development and personal use most

play01:14

likely if you're going to run it on a

play01:15

home computer this is the model that

play01:17

you're going to use this model is also

play01:19

openly available under the Apache 2.0

play01:22

license so it is open source any tools

play01:25

that you create that actually use flux

play01:26

one Schnell underneath can be sold any

play01:30

images that were generated with this

play01:31

tool can also be used non-commercially

play01:34

or commercially however you want then

play01:36

you have the next model up flux one Dev

play01:39

this is their middleof thee line model

play01:41

it's slightly more efficient than their

play01:43

topof the line model and slightly more

play01:45

prompt adherent and performant than

play01:48

their flux one Snell model now this one

play01:50

can be used for non-commercial

play01:53

applications so you can't create a tool

play01:55

that's using this and then sell access

play01:58

to that tool and then you've got their

play01:59

topthe line model Flex one pro which is

play02:02

their best model it offers their

play02:04

state-of-the-art performance and is

play02:06

really designed for Enterprise Solutions

play02:08

now if you want to use one of these flux

play02:10

models there's actually a handful of

play02:12

websites that have already integrated it

play02:14

and you can use it completely for free

play02:16

right now probably the simplest way to

play02:18

use it is to head on over to Black

play02:20

Forest Labs over on hugging face and you

play02:23

can actually use their Snell model their

play02:26

lowest end model and their Dev model

play02:28

within hugging face spaces so if I click

play02:31

into this space here I get a real basic

play02:35

image generator for flux one Schnell I

play02:38

jump back here you can see we've also

play02:39

got flux one Dev model both of them are

play02:42

just pretty Bare Bones you enter your

play02:44

prompt we can see where it generates the

play02:46

image and we have a few advanced

play02:48

settings like a random seed with height

play02:51

and number of inference steps if you do

play02:53

want to use the pro model there's

play02:55

actually a platform called glyph which

play02:57

is this really cool AI workflow flow

play03:00

Builder and you can build your own flux

play03:03

workflows in here and it will let you

play03:05

generate for free even using the pro

play03:07

model so if I click on build up here I

play03:10

can start building with glyph blocks we

play03:12

want to start with a text input here

play03:15

what is your prompt if we want we can

play03:17

even run that prompt through an llm like

play03:21

clad or chat GPT to improve the prompt

play03:24

so let's retitle this from input one to

play03:27

basic prompt let's go ah head and run it

play03:31

through a text generator let's tell it

play03:33

to take the following image prompt and

play03:35

prove it so that it generates colorful

play03:37

high contrast images and then I'll put

play03:39

quotation marks we'll put our basic

play03:41

prompt in there close the quotation

play03:43

marks under Advanced controls let's go

play03:46

ahead and set the model to Cloud 3.5

play03:49

Sonet we'll rename this one from text

play03:51

one to optimized prompt and now we can

play03:55

add another block to generate an image

play03:57

so let's go ahead and create our image

play03:59

generator here and we'll go ahead and

play04:01

have it use our optimized prompt here

play04:04

under image generation model you can see

play04:06

we have the option for flux Pro and flux

play04:08

Schnell so let's go ahead and use flux

play04:10

Pro we've got nothing to lose right now

play04:13

cuz glyph is letting you actually use

play04:14

this model for free let's go ahead and

play04:17

do a landscape 16x9 image here now we

play04:21

can go ahead and close all of these down

play04:24

and now it asks what is your prompt and

play04:26

what it will do is take the basic prompt

play04:28

run it through cloud 3.5 Sonet improve

play04:31

it for contrast and colorfulness and

play04:33

then output an image for us and let's go

play04:36

ahead and just do a wolf howling at the

play04:38

moon and run this glyph you can see it's

play04:40

looking at my basic prompt and then it's

play04:43

optimizing that prompt and now it's

play04:45

generating an image based on that

play04:46

optimized prompt and here's the image

play04:49

that it generated pretty dang solid

play04:51

honestly I can actually open this step

play04:54

over on this left sidebar and see what

play04:56

the optimized prompt was you can see it

play04:59

is a majestic wolf with lustrous silver

play05:01

and charcoal fur standing at top a

play05:03

rugged Cliff howling at an enormous

play05:05

luminous full moon the night sky is a

play05:08

vibrant mix of deep Indigo and swirling

play05:10

purple clouds etc etc and you can see it

play05:12

got all of that into this now I

play05:15

personally haven't spent a ton of time

play05:18

playing around with flux myself but a

play05:20

good buddy of mine named Miguel who goes

play05:22

by Angry penguin PNG over on Twitter he

play05:26

has spent a lot of time with it and if

play05:27

you're not following angry penguin I

play05:29

highly recommend you follow him on

play05:30

Twitter he is at The Cutting Edge of

play05:32

everything that's going on with AI a lot

play05:34

of the information that I get before a

play05:36

lot of people get it I get from AP over

play05:40

here on X so make sure you're following

play05:41

him if you're not but I called him up

play05:45

because I wanted to get some details

play05:46

from him about what flux is good at what

play05:49

flux isn't great at and what other

play05:51

details we should actually know when

play05:54

going into use this tool so let's start

play05:56

with what flux isn't great at it's not

play05:59

good at illustrations from what I've

play06:01

seen like it can be better but I've seen

play06:03

like when people are doing like the fine

play06:04

tunes of like their sdxl they can get

play06:06

much nicer things for illustrations so

play06:08

let's try an illustration we'll go with

play06:10

a handdrawn illustration of an angry

play06:12

penguin and this actually looks like a

play06:15

pretty good image but it doesn't really

play06:17

look handdrawn in my opinion so I can

play06:19

see where maybe it's missing some of

play06:21

that detail let's try an oil painting of

play06:24

an angry penguin and once again it looks

play06:26

pretty good but it doesn't scream oil

play06:29

painting to me all right last one let's

play06:31

try a watercolor painting of an angry

play06:33

penguin yep that looks like an angry

play06:35

penguin to me but it's got some

play06:37

watercolor elements in the background

play06:39

and maybe some splashes but definitely

play06:41

the penguin itself is not looking like a

play06:44

watercolor to me not bad but if I look

play06:46

at what mid Journey generates with the

play06:48

same prompt a watercolor painting of an

play06:50

angry penguin it definitely screams

play06:53

watercolor to me a little bit more or if

play06:55

I ask it to generate an oil painting of

play06:58

an angry penguin defin itely looks more

play07:00

oil painting like you definitely get

play07:02

those brush Strokes like you'd see in in

play07:04

oil painting I didn't see that as much

play07:07

when using the flux one model so I went

play07:10

ahead and published my glyph real quick

play07:12

and you can see it actually gave me a

play07:13

different user interface but now let's

play07:15

talk quickly about what flux is really

play07:18

good at is it designed to do realism

play07:21

really well yeah I would say that the

play07:23

team did a lot of work on aesthetic

play07:25

training right and then just making sure

play07:27

that it produces really great quality

play07:28

outputs like right out of the box I

play07:30

would say it's on par with mid Journey

play07:32

if you're using the right prom so let's

play07:34

test a photo realistic image let's try a

play07:37

photo of a man with a beard on a city

play07:39

sidewalk eating an ice cream and it's

play07:41

pretty darn good I would say our dude

play07:43

might be sniffing the ice cream more

play07:45

than actually eating it but it's pretty

play07:47

realistic looking let's go ahead and try

play07:49

another prompt and let's do a woman

play07:51

taking a selfie on a tropical island

play07:53

with the beach in the background and

play07:54

let's see what we get now that's not bad

play07:56

at all I can tell it's AI it doesn't

play07:58

look Ultra realistic to me but that

play08:00

could also be the prompt optimization

play08:02

that I put in let's go ahead and build a

play08:04

even simpler workflow where we just

play08:06

enter our prompt here and then just

play08:08

generate the image so we'll just put

play08:10

prompt and we'll have it just use our

play08:13

input to generate the image here and

play08:15

we'll use flux Pro and now we have a

play08:17

very very basic image generator I'll go

play08:19

ahead and publish it so we get that

play08:21

other user interface and now it's not

play08:23

going to optimize my prompts this is

play08:24

going to come straight out of the image

play08:26

generator let's do a portrait of a woman

play08:28

in a corporate setting about to give a

play08:29

meeting and we can see it generated a

play08:31

pretty realistic image now when I use

play08:34

this same a portrait of a woman in a

play08:35

corporate setting about to give a

play08:37

meeting I feel like mid Journey still

play08:39

has it beat by a little bit on the

play08:42

realism here these images look much more

play08:44

realistic much more like something you'd

play08:46

actually find on a stock photo website

play08:48

but they also kind of look the same

play08:50

there's nothing that says about to give

play08:52

a meeting in any of these images where

play08:54

the glyph version she's clearly about to

play08:56

give a meeting what flux is really great

play08:58

at is do anything that has to do with

play09:00

text so you can make like logos with it

play09:02

they come out like insane you can create

play09:04

like Snapchat selfies that look real I

play09:06

was doing like an example where it was

play09:08

like George Washington crossing the

play09:10

Delaware and it has a selfie and it came

play09:12

out like super insane just like little

play09:13

things like that I think it's going to

play09:14

be really great for making memes the

play09:16

text alone is something that we did not

play09:18

have with sd3 we really didn't have it

play09:20

with sdxl Dolly 3 kind of does text

play09:22

pretty well but Lux is definitely a step

play09:24

above so let's do a polar bear holding a

play09:27

sign that says Mr e flow and it pretty

play09:29

much nailed it on the first try I don't

play09:31

see anything wrong with that let's try

play09:33

this one a plane writing the word

play09:35

subscribed to Matt wolf out of smoke in

play09:37

the sky let's see how it does with a

play09:38

little bit longer text I mean not

play09:40

horrible it says subscribe Wolfie it's

play09:43

almost there but it definitely sort of

play09:46

lost itself a little bit on the longer

play09:49

text here all right so running this

play09:51

prompt one more time this time it just

play09:53

got the word subscribe but none of the

play09:55

rest of it I think the plane looks a lot

play09:57

better on this one yeah I would say

play09:59

prompt adherence is also a huge one

play10:01

aesthetic quality the fact that it's

play10:02

like not censored right is like a huge

play10:04

thing as well for people that are

play10:05

creating locally it is censored like it

play10:07

does have an NSFW filter so it's

play10:09

centered in the good ways but not in the

play10:10

ways that you know you want to create

play10:12

like maybe a political figure just for

play10:13

fun or something like that and you won't

play10:14

get hit with like a shadow ban or

play10:16

something like that so I think just

play10:17

having that optionality is pretty

play10:19

important as well now this one's also

play10:21

supposedly really good at prompt

play10:22

adherence an area that mid journey is

play10:24

not great at and when I say prompt

play10:26

adherence I mean adding a lot of things

play10:28

into the prompt and having it get all of

play10:30

those elements so for example a

play10:33

three-headed dragon watching TV while

play10:34

eating nachos and wearing a cowboy boots

play10:37

if I was to plug this into mid Journey

play10:39

I'd likely get a couple of the elements

play10:40

I'd probably get like the three-headed

play10:42

dragon and maybe a TV but it would miss

play10:44

the nachos and the cowboy boots watch

play10:46

one area that Dolly 3 is really good at

play10:48

it captures all of the stuff from your

play10:50

prompt and usually figures out how to

play10:51

get it all in there so let's see how

play10:54

flux one does with getting as many

play10:57

elements into the prompt as possible so

play10:59

definitely not bad I wouldn't say this

play11:01

is a three-headed dragon this looks more

play11:03

to me like two separate dragons but it

play11:05

got the nachos it got the TV it got the

play11:07

cowboy boots let's give it one more go

play11:10

around this time it looks more like a

play11:11

one-headed dragon eating spaghetti

play11:14

watching TV without the cowboy boots so

play11:17

I mean it's prompt adheren is but I'd

play11:20

probably put it about the same level as

play11:22

mid Journey when plugging it into mid

play11:24

Journey here's what I got one with three

play11:26

dragons eating nachos no cowboy boots

play11:29

here's one with just a normal Dragon

play11:30

wearing cowboy boots eating chips but no

play11:32

TV here's three dragons eating burritos

play11:36

or something and here's three dragons

play11:38

eating chips none of them got the threee

play11:39

headed dragons with the cowboy boots

play11:41

watching TV eating nachos none of them

play11:43

got them all right now compare that to

play11:45

what we get when we plug it into Dolly 3

play11:48

check this out three-headed dragon

play11:50

eating nachos wearing cowboy boots

play11:52

watching TVs even got a bowl that says

play11:55

nachos on it so Dolly 3 when it comes to

play11:58

prompt adherence

play11:59

absolutely crushes both flux one and mid

play12:03

Journey but Dolly 3 kind of still sucks

play12:05

at realism now the other thing I want to

play12:07

test is the sort of uncensored of flux

play12:11

now you can't generate not safe for work

play12:13

stuff yet it's open source people are

play12:15

eventually going to break that and if

play12:17

you want to generate not safe for work

play12:19

stuff it's only a matter of time but

play12:21

right now you can't do that but

play12:23

theoretically you can generate

play12:25

copyrighted images and use existing IP

play12:28

and I want to kind of test that let's do

play12:30

SpongeBob SquarePants high-fiving Super

play12:32

Mario and I mean the hands sort of merg

play12:36

together so that's not great but we

play12:37

definitely got SpongeBob and we

play12:39

definitely got Mario let's see what

play12:41

happens when we try to get like

play12:42

celebrities into it let's do Tom Hanks

play12:45

hugging Kanye West I'd say that looks

play12:47

more like Kanye West hugging an

play12:49

alternate version of Kanye

play12:52

West but that definitely looks like

play12:54

Kanye let's go ahead and do um Tom Hanks

play12:57

standing next to Kanye that did a pretty

play13:00

good job the Kanye looks pretty spoton

play13:03

Tom Hanks not quite there but the point

play13:06

being it's not going to disallow you to

play13:10

generate whatever you want if I want

play13:13

Donald Trump eating a donut I can do

play13:15

that if I want kamla Harris eating tacos

play13:18

I can do that and if I want an image of

play13:20

baby Yoda hanging out with Spider-Man I

play13:22

can also do that if you want some

play13:25

additional tips for how to improve your

play13:27

prompts with this a you you can use an

play13:30

AI method where you give it a basic

play13:32

prompt and then AI automatically

play13:33

improves The Prompt for you or here's

play13:36

some tips directly from Miguel let's say

play13:37

somebody just wants to mess with

play13:38

prompting themselves like without using

play13:40

the prompt tuner like what are some of

play13:42

the things you've noticed work well cuz

play13:43

I remember like early on with stable

play13:45

diffusion if you added things like 4K

play13:47

and HD or octane render or Unreal Engine

play13:51

or you know you added some of these

play13:52

little keywords to the end it would

play13:54

actually have a pretty big impact on the

play13:56

output image is there any like little

play13:58

tricks or things you've noticed to

play14:00

actually get better outputs from the

play14:02

images yeah so it does have a dual

play14:04

encoder so you could use like the simple

play14:05

prompts and the more complex prompts but

play14:07

if you really want like something super

play14:09

like insane I would say try and be as

play14:11

detailed with the prompt as possible

play14:13

fler who puts amazing stuff on X they

play14:16

are like always pushing the boundaries

play14:17

of the new and upcoming models they have

play14:18

like some insane prompts that they're

play14:20

putting in like very descriptive down to

play14:21

the word like spaghetti monster coming

play14:23

out of a donut and you know attacking

play14:25

something and it's just like super

play14:26

descriptive super well like I'm looking

play14:28

at right now that's like the word cat

play14:30

where the shapes of all the letters

play14:32

combined to form a cat using only clever

play14:34

typography and it's like spelling out

play14:36

the word cat with like a cat in there

play14:38

right it's like insane what you can do

play14:39

just with you know really using your

play14:41

imagination to get something creative

play14:44

now one thing that's really cool about

play14:45

this flux model is they've already

play14:48

announced down here that this is also

play14:50

going to be a text to video model so

play14:52

right now flux one is text to image but

play14:56

they will serve as the powerful

play14:58

foundation for the their upcoming Suite

play14:59

of competitive generative text to video

play15:02

systems so pretty soon using flux one

play15:06

you'll have an open-source option to

play15:09

generate the types of things that we've

play15:10

seen from tools like lum's dream machine

play15:13

Runway gen 3 Sora which nobody really

play15:16

still has access to pretty soon you'll

play15:18

have an open-source version to do that

play15:21

with flux one now does flux one really

play15:24

kill mid Journey like some of the

play15:26

headlines are saying not yet mid journey

play15:28

is still czy Superior images Dolly 3

play15:31

still has much better prompt adherence

play15:33

but flux one is definitely better than

play15:35

what we're getting out of the existing

play15:37

stable diffusion models like sdxl and

play15:39

stable diffusion 3 and they're close

play15:42

close to on par with what we're getting

play15:44

out of mid journey in terms of realism

play15:48

text generation and getting a lot closer

play15:51

to the prompt adherence that we're

play15:52

seeing from Dolly 3 so this could if it

play15:55

continues to evolve which of course it

play15:57

will be a content

play15:59

to do all of the things that mid Journey

play16:01

does well Dolly 3 does well and stable

play16:04

diffusion does well Dolly 3 great at

play16:06

prompt adherance mid-journey great at

play16:08

realism stable diffusion great at pretty

play16:11

much uncensored anything you can

play16:13

possibly dream up in your mind you can

play16:16

do it without any guard rails make

play16:18

whatever you want flux looks like it

play16:20

could be the best of all three of those

play16:22

worlds soon keep in mind that the flux

play16:25

Schnell model is open source so other

play16:28

developers are going to get their hands

play16:30

on it they're going to fine-tune it

play16:31

they're going to improve it they're

play16:32

going to rip off the few guard rails

play16:35

that it has now people are going to

play16:36

create luras for these and you're going

play16:38

to be able to generate better and better

play16:40

images with it all with an open model

play16:43

that's free to use that's open source

play16:45

that you should be able to download on

play16:46

your computer pretty soon and run

play16:49

locally if you wanted to so we're not

play16:51

far off from being able to just install

play16:53

this on our own computers run them

play16:55

without being connected to the internet

play16:57

and generate anything you can imagine so

play17:00

I'm pretty excited about it personally

play17:02

I'm likely still going to go to Leonardo

play17:05

or mid joury or Dolly 3 depending on my

play17:09

use case but I would say give it a month

play17:11

or so and this flux model and whatever

play17:15

people build on top of it is probably

play17:18

going to be generating much better

play17:19

images than those platforms so exciting

play17:23

exciting advancement in the world of AI

play17:24

art I'm super pumped to play around with

play17:27

it more if I find even more cool tips

play17:29

and tricks and ways to improve my

play17:32

outputs I'll make a follow-up video to

play17:33

this one but right now I'm just excited

play17:35

that there's a new open source model to

play17:37

compete with stable diffusion that

play17:40

already seems to be better than stable

play17:42

diffusion again I'll keep you looped in

play17:44

as I learn more thank you so much for

play17:46

tuning in thank you once again to Miguel

play17:48

AKA angry penguin for helping me out

play17:50

with this video and walking me through

play17:52

what flux is and isn't good at if you

play17:54

want to stay looped in with all the

play17:56

latest cool AI tools and news check out

play17:58

future tools. and join the free

play18:00

newsletter I'll send you an email to

play18:02

your inbox with just the coolest tools

play18:03

and most important news each week and if

play18:06

you like videos like this you want to

play18:08

see more tutorials news and breakdown of

play18:12

the latest happenings in the AI World

play18:14

make sure you like this video And

play18:15

subscribe to this channel I'll make sure

play18:17

stuff like this keeps on showing up in

play18:18

your YouTube feed thank you once again

play18:20

for nerding out with me today really

play18:22

appreciate you see you in the next one

play18:23

bye-bye

Rate This

5.0 / 5 (0 votes)

الوسوم ذات الصلة
AI-BildgenerierungFlux OneBlack Forest LabsRealismusText zu BildOpen SourceAI-TechnologieInnovationBildbearbeitungDigital Kunst
هل تحتاج إلى تلخيص باللغة الإنجليزية؟