ComfyUI FLUX IMAGE TO IMAGE FLORENCE 2 workflow #comfyui #flux #img2img #florence

FLUX IMAGE TO IMAGE FLORENCE 2 workflow
introduction to FLUX model, a new and super interesting model that knows how to relate very well to our prompts .
And also in terms of quality, it gives competition to Midjourney and Dall e.

In addition, we will see an image to image workflow that uses the ability of florence2 to describe our images, and improve them significantly with the help of flux
We will work with the version flux schnell wich is openly available under an Apache2.0 license

#comfyui #flux #blackforestlabs #img2img #aiphotography #florence

Previous video explaining Florence captioning

workflow

florence 2 repo

flux text encoder (t5xxl + clip l )

flux vae

flux schnell model

black forest labs

ComfyUI flux example

[pauloparanhos]

🎯 Key points for quick navigation:

00:00:00 *🌟 Introdução ao modelo Flux e tipos de licenças*
- Introdução ao modelo Flux e suas capacidades em relação às prompts.
- Flux oferece três tipos de licenças: Flux Pro, Flux Dev e Flux Shell, cada uma com suas permissões específicas.
00:02:02 *🖼️ Workflow básico de texto para imagem com Flux*
- Demonstração da configuração básica para usar o Flux no processo de texto para imagem.
- Requisitos como instalação de modelos específicos e organização de arquivos.
- Exemplos de prompts e resultados obtidos, destacando a qualidade e fidelidade às descrições.
00:06:34 *📸 Workflow de imagem para imagem com Flux e Florence 2*
- Explicação detalhada do workflow para transformar imagens usando Flux e Florence 2.
- Utilização do denoise para melhorar a qualidade das imagens geradas.
- Exemplos práticos de como as imagens originais são processadas e melhoradas pelo modelo.

Made with HARPA AI

[Lexie-bqkk]

"I don't know if you need it" Thank you, better than some you tubers who just say "This is this best setting, use this" and they don't actually know how or why that's the case, just some anecdote they stumbled upon.

[c_huntermc]

The thing I'm most excited about isn't just the image quality, it's the more descriptive prompting. I'm already quite happy with SD 1.5 image quality, but usually it just seems to take elements from your prompt and just mix them up in random order.

[sinayagubi]

Apache 2 allows commercial use. I am amazed people in the image generation niche aren't familiar with this license. The only thing you can't do is pretending that the makers of flux are endorsing you.

Thank you for this amazing video. I learned a lot. ♥
BTW 8:27 just personal opinion, but I prefer the image on the right much more than the 0.73 denoise one. I laughed when you said "much more atractive" while I was visually presented with opposite :D

Great! Special thanks for introducing the Florance2 option!

[swannschilling]

I did tune in because of Florence...happy to see that its also Flux!! 😊

[trollenz]

This workflow is the 💣 ! Thanks mate 👌🏻

[pneydny]

A seriously underrated content creator - supporting you into the 100k+

Genuiinely helpful!

[DarkionAvey]

really appreciate the comfyui community. thanks

[taucalm]

Quality seems to be a little better than sdxl, but we have no idea if the model is better or worse because its almost 4 times size of sdxl.

[ZakariaNada]

the original ones look better and realistic, thanks for the video

[sertocd]

I found out that even low number 4 steps, uni_pc_bh2 for the sampler and sgm_uniform scheduler is the best for maintaining fingers and toes appear the way they are supposed to.

[VintageForYou]

Great video on using FLUX I would love you to make more videos on this amazing AI FLUX Tool.🙋🏼‍♂

[MichaelUlstrup]

hey, thanks for the great video, Florence 2 is giving me some problems, it tells me to install torch 2.1.1 but i have version: 2.4.0, i think ...

[JuneStarling]

Super helpful! Florence reminds me of CGDream

[monkehh]

do u know how to create same face and clothes of a model ? or create same model in diff position

[bilalalam]

I am switching from automatic111 to comfyui - thank you for vidoe - and keep making more

[testales]

The Flux workflow is overcomplicated. You can simply use a normal KSampler and wire it as usual, it will you exactly the same results as long as you stick to CFG 1. This CFG 1 is also one of the big issues with Flux as it means your negative prompt ist ignored. I wonder if they provided this flow to hide this fact. Anyway, if you wire it the usual way, you can at least slightly increase the CFG, 1.5 looks still good in some cases even a little better but it quickly gets worse. But that way you can at least have a somewhat working negative prompt. Btw. isn't 4 steps a little too low?

[DJVARAO]

How much VRAM an RAM you used? how long it takes i2i?