license: other
license_name: flux-1-dev-non-commercial-license
license_link: >-
https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/blob/main/LICENSE.md
base_model:
- black-forest-labs/FLUX.1-dev
====================================
FLUX Model Merges & Tweaks: Detail Enhancement and Acceleration
| Detail enhancement and acceleration techniques have been applied, particularly on NVIDIA 4XXX cards (maybe 3XXX too). The goal is efficient acceleration with lower overhead. Distilled models can be used with CFG as a convenient bonus. Baked-in accelerators are reliable, except maybe Turbo – it's still finding its footing. As for updates, they might arrive in a flurry or after a long hiatus – nobody can predict the workflow whims of fate. NOTE: These are by-products from workflow optimizations. Things can change at any given time for no apparent reason. Shit can appear and vanish, particular file types, may or may not be listed, etc.. |
|
====================================
Detail Plus! - Built upon the base model Freepik's Flux.1-Lite-8B-alpha:
Detail Enhancement Used:
- Style LORA - Extreme Detailer for FLUX.1-dev (Weight: 0.5) (Model Link)
- Best of Flux: Style Enhancing LoRA (Weight: 0.25) (Model Link)
SAFETensors Format (fp8_34m3fn_fast):
GGUF Quantized Models (Q8_0):
====================================
Detail Plus! Distilled - Built upon the base model Freepik's Flux.1-Lite-8B-alpha:
Detail Enhancement Used:
- Style LORA - Extreme Detailer for FLUX.1-dev (Weight: 0.15) (Model Link)
- Best of Flux: Style Enhancing LoRA (Weight: 0.06) (Model Link)
Distillation Used:
- Flux distilled lora (Weight: Hyper - 1.00 | Turbo - 0.50) (Model Link)
- GGUF Quantized Models (Q8_0):
====================================
Detail Plus! De-Re-Distilled (Built upon the base model Flux-dev-de-distill:
Detail Enhancement Used:
- Style LORA - Extreme Detailer for FLUX.1-dev (Weight: 0.15) (Model Link)
- Best of Flux: Style Enhancing LoRA (Weight: 0.15) (Model Link)
Re-Distillation Used:
- Flux distilled lora (Weight: -1.00) (Model Link)
SAFETensors Format V2 (fp8_34m3fn_fast):
GGUF Quantized Models (Q8_0):
GGUF Quantized Models (Q6_K):
GGUF Quantized Models (Q4_K_S):
====================================
Acceleration Credits:
ByteDance, a leading AI technology company, optimized the following models:
Alimama Creative, a renowned NLP innovator, optimized the following model:
====================================
Attribution and Licensing Notice:
The FLUX.1-dev Model is licensed by Black Forest Labs, Inc. under the FLUX.1-dev Non-Commercial License. Copyright Black Forest Labs, Inc.
Our model weights are released under the FLUX.1-dev Non-Commercial License.
This merge combines the strengths of these models, applying detail enhancement and acceleration techniques to create a unique and powerful AI model built upon Freepik's Flux.1-Lite-8B-alpha & Flux-dev-de-distill. We hope this contributes positively to the development of the NLP community!