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!