FLUX KONTEXT OPEN SOURCE! 8GB Vram Local Generation!

The time has come ladies and gentleman - Flux Kontext has gone open-source! With some caveats... The full model is fairly hefty to run rough

The time has come ladies and gentleman - Flux Kontext has gone open-source! With some caveats...

The full model is fairly hefty to run roughly 32-80gb of Vram required! That's why we've found quantized versions for you, uploaded just hours before this post!

We tested the Q5 version on a 3060ti (8GB Vram) and saw pretty solid results with fairly quick generation speeds (<3min).

While obviously higher quantized models will be better, it's ultimately a battle between generation speed and quality.

The open source Flux Kontext model also supports Lora's which has been integrated into the attached workflow. Documentation also states that ControlNet is supported, though I haven't tested this yet. Curious to hear results from anyone who does!

Ultimately, these models becoming open-source is great news! It opens the door for selective editing using masks and segment anything models. Excited to see the workflows that get cooked up in the coming weeks - we have a couple ideas of our own!

Resources:

BFL Flux Kontext Dev Full Model Hugging Face: https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev

Flux Kontext Dev Quantized: https://huggingface.co/bullerwins/FLUX.1-Kontext-dev-GGUF

Flux Kontext Local Reddit Thread: https://www.reddit.com/r/LocalLLaMA/comments/1ll38zu/flux1_kontext_dev_an_open_weights_model_for/

Workflows attached! Would love to see what you guys create!