File size: 5,065 Bytes
4966da1 42217f6 530b61a 4966da1 a8ca6b8 271bb72 4966da1 58fad84 9489fe0 271bb72 9489fe0 f8143b0 271bb72 9489fe0 4966da1 42217f6 4966da1 42217f6 4966da1 271bb72 d34432f 271bb72 4966da1 271bb72 4966da1 d34432f 4966da1 a8ca6b8 4966da1 a8ca6b8 4966da1 42217f6 4966da1 a8ca6b8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 |
---
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**
=====================================================
<img src="https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/resolve/main/images/ComfyUI_2024-11-24_0231.jpg" alt="ComfyUI 2024-11-24 0231">
This repository contains merged models, built upon the base models:
- [Freepik's Flux.1-Lite-8B-alpha](https://huggingface.co/Freepik/flux.1-lite-8B-alpha)
- [Flux-dev-de-distill](https://huggingface.co/nyanko7/flux-dev-de-distill)
Detail enhancement and acceleration techniques have been applied, particularly optimized for NVIDIA 4XXX cards (maybe 3XXX too). The goal is to have high efficiency accelerated models with lower overhead. The (de-re-destill) model can be used with CFG remaining at 1. Also the baked-in accelerators work as intended.
**Detail Plus! - Built upon the base model [Freepik's Flux.1-Lite-8B-alpha](https://huggingface.co/Freepik/flux.1-lite-8B-alpha):**
**Detail Enhancement Used:**
- **Style LORA - Extreme Detailer for FLUX.1-dev** (Weight: 0.5) ([Model Link](https://civitai.com/models/832683))
- **Best of Flux: Style Enhancing LoRA** (Weight: 0.25) ([Model Link](https://civitai.com/models/821668))
1. **GGUF Quantized Models (Q8_0)**:
- [flux.1-lite-8B-alpha-Hyper-16.Steps-Detail.Plus.Steps.Q8_0_quantized](https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/blob/main/flux.1-lite-8B-alpha-Hyper-16.Steps-Detail.Plus.Steps.Q8_0_quantized.gguf)
- [flux.1-lite-8B-alpha-Hyper-8.Steps-Detail.Plus.Steps.Q8_0_quantized](https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/blob/main/flux.1-lite-8B-alpha-Hyper-8.Steps-Detail.Plus.Steps.Q8_0_quantized.gguf)
- [flux.1-lite-8B-alpha-Turbo-8.Steps-Detail.Plus.Steps.Q8_0_quantized](https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/blob/main/flux.1-lite-8B-alpha-Turbo-8.Steps-Detail.Plus.Steps.Q8_0_quantized.gguf)
2. **SAFETensors Format (fp8_34m3fn_fast)**:
- [flux.1-lite-8B-alpha-Hyper-16.Steps-Detail.Plus-fp8_e4m3fn_fast](https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/blob/main/flux.1-lite-8B-alpha-Hyper-16.Steps-Detail.Plus-fp8_e4m3fn_fast.safetensors)
- [flux.1-lite-8B-alpha-Hyper-8.Steps-Detail.Plus-fp8_e4m3fn_fast](https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/blob/main/flux.1-lite-8B-alpha-Hyper-8.Steps-Detail.Plus-fp8_e4m3fn_fast.safetensors)
- [flux.1-lite-8B-alpha-Turbo-8.Steps-Detail.Plus-fp8_e4m3fn_fast](https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/blob/main/flux.1-lite-8B-alpha-Turbo-8.Steps-Detail.Plus-fp8_e4m3fn_fast.safetensors)
**Detail Plus! De-Re-Distilled (Built upon the base model [Flux-dev-de-distill](https://huggingface.co/nyanko7/flux-dev-de-distill)
**Detail Enhancement Used:**
- **Style LORA - Extreme Detailer for FLUX.1-dev** (Weight: 0.15) ([Model Link](https://civitai.com/models/832683))
- **Best of Flux: Style Enhancing LoRA** (Weight: 0.15) ([Model Link](https://civitai.com/models/821668))
**Re-Distillation Used:**
- **Flux distilled lora** (Weight: 0.15) ([Model Link](https://civitai.com/models/977247/flux-distilled-lora))
- Flux-dev-de-re-distill-Hyper.16.Step-Detail.Plus-ReDis-fp8_e4m3fn_fast
- Flux-dev-de-re-distill-Hyper.8.Step-Detail.Plus-ReDis-fp8_e4m3fn_fast
- Flux-dev-de-re-distill-Turbo.8.Step-Detail.Plus-ReDis-fp8_e4m3fn_fast
**Acceleration Credits:**
* [ByteDance](https://huggingface.co/ByteDance), a leading AI technology company, optimized the following models:
- [Hyper-FLUX.1-dev-16steps](https://huggingface.co/ByteDance/Hyper-SD/blob/main/Hyper-FLUX.1-dev-16steps-lora.safetensors)
- [Hyper-FLUX.1-dev-8steps](https://huggingface.co/ByteDance/Hyper-SD/blob/main/Hyper-FLUX.1-dev-8steps-lora.safetensors)
* [Alimama Creative](https://huggingface.co/alimama-creative), a renowned NLP innovator, optimized the following model:
- [FLUX.1-Turbo-Alpha](https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha)
**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](https://huggingface.co/mdmachine/FLUX.Model.Merge-Detail.Enhancement.and.Acceleration/blob/main/LICENSE.md).
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](https://huggingface.co/Freepik/flux.1-lite-8B-alpha). We hope this contributes positively to the development of the NLP community! |