Hyper Flux1 dev NF4 checkpoint for 8 steps inference

[WindyYam]

In case no one post before.
Just convert the merged version of flux1 dev + hyper flux lora into nf4, which has speed and size advantage together, now it can also generate decent good result in 8 steps
https://huggingface.co/ZhenyaYang/flux_1_dev_hyper_8steps_nf4/tree/main

I have a very low VRAM laptop (4GB), and this is my savior for now (around 1 min at 1152x922) before I get a new laptop

[BSG-Walter]

thanks, i was about to convert it for myself the other day, but got distracted with something.
downloading...

[ErmMonzon]

Amazing! Thank you!

[dongxiat]

what is ur card... im 4GB 1050ti around 7~10min?

[WindyYam]

A nVidia A500 4GB, I think it is just slightly better than 1050ti, compare
what's your configuration? Did you check the low bit diffusion as automatic or nf4? Also, 8 steps is enough

[Moussack]

hey how did you convert checkpoint to NF4 version??

[WindyYam]

The latest sd forge has this feature already in the checkpoint merger tab

[Moussack]

So is this the tool that you talk about??

[wisdombox44]

Works great and loras works fine thanks :)

[AgathoklisICXC]

Can you provide the workflow?

[WindyYam]

Na, I suppose everyone is using sd forge here? I haven't used ComfyUI yet, as sd forge is rather easy to use for a noob like me~~~

[suonnon]

Can I use the images which were generated by this model for commercial purposes?

[Giribot]

Yes for Flux.1-Schnell, no for "Flux.1-dev".
Strange no ?
Source: https://civitai.com/articles/6625/can-i-use-flux-for-commercial-use

[suonnon]

@Giribot and it also applies to version "hyper" ?

[timwillhack]

actually... you can use the IMAGES for commercial purposes: "Outputs. We claim no ownership rights in and to the Outputs. You are solely responsible for the Outputs you generate and their subsequent uses in accordance with this License. You may use Output for any purpose (including for commercial purposes), except as expressly prohibited herein. You may not use the Output to train, fine-tune or distill a model that is competitive with the FLUX.1 [dev] Model."

[danilomaiaweb]

This checkpoint is really very good. I tested it today and was surprised. Look...

Prompt:
A charismatic speaker is captured mid-speech. He has long, slightly wavy blonde hair tied back in a ponytail and is adorned with a prominent, well-groomed salt-and-pepper beard and mustache. The beard is full and neatly styled, while the mustache is bold and meticulously shaped, drawing attention as part of his striking appearance. His expressive face, complete with a confident smile, is animated as he gestures with his left hand, displaying a large ring on his pinky finger. He is holding a black microphone in his right hand, speaking passionately. The man is wearing a dark, textured blouse with unique, slightly shimmering patterns, and a green lanyard with multiple badges and logos hanging around his neck. The lanyard features the 'Autodesk' and 'V-Ray' logos prominently. Behind him, there is a blurred background with a white banner containing logos and text, indicating a professional or conference setting. The overall scene is vibrant and dynamic, capturing the energy and charisma of a live presentation

Steps: 10
Sampler: [Forge] Flux Realistic
Schedule type: Simple
cfgScale: 1
Distilled CFG Scale: 3.5
Size: 768x1024
Model: Flux1-DEV-Hyper-8STEPS-NF4

This image was generated in only
Time taken: 47.6 sec.

My Setup:
AMD Ryzen 5 3500X 6-Core Processor - 3.59 GHz
32,0 GB Ram
NVIDIA GeForce RTX 3050 - 8GB
Win 11
My Forge: version: f2.0.1v1.10.1-previous-635-gf5330788