Hugging Face's logo

ND911
/
EclecticEuphoria_Flux_Unet_Schnell_models

GGUF
Model card Files Community
EclecticEuphoria_Flux_Unet_Schnell_models
File size: 6,530 Bytes 
27817a6
47fce32
 
27817a6
 
c16e3df
 
 
 
 
 
 
 
 
 
4c993c7
 
 
 
 
 
 
 
0926cd9
 
4c993c7
c16e3df
1d4de15
47fce32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1f8cdd7
 
47fce32
 
 
 
8361cfe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24b18fe
 
 
 
 
 
 
 
 
10ecda0
 
 
 
8361cfe
 
27817a6
11faf4b
27817a6
 
 
 
 
 
 
 
 
 
 
 
48ea2a7
27817a6
48ea2a7
27817a6
48ea2a7
 
 
 
f97c939
 
 
 
48ea2a7
 
47fce32
 
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
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
 
---
base_model:
- black-forest-labs/FLUX.1-schnell
---

# EclecticEuphoria_Flux_Unet_Schnell_v4e 

[EclecticEuphoria_Flux_Unet_Schnell_v4e-f16.safetensors](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v4e_f16.safetensors?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v4e-Q8_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v4e_Q8_0.gguf?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v4e-Q6_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v4e_Q6_K.gguf?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v4e-Q4_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v4e_Q4_K.gguf?download=true)


Apparently v3e has nipple control issues, as in coming through clothing, v4e should fix that

  * ![](xcompare.png)


v4 is an experiment, thanks to a new node [FluxBlockPatcherSampler](https://github.com/cubiq/Block_Patcher_ComfyUI)

[Civitai Gallery of Schnell Images from this model](https://civitai.com/collections/1216945)

------------------------------------------------------------------------------------------------------------------------------

# EclecticEuphoria_Flux_Unet_Schnell_v3e 

[EclecticEuphoria_Flux_Unet_Schnell_v3e-f16.safetensors](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v3e_f16.safetensors?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v3e-Q8_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v3e_Q8_0.gguf?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v3e-Q6_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v3e_Q6_K.gguf?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v3e-Q4_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v3e_Q4_K.gguf?download=true)

v3 is an experiment, thanks to a new node [FluxBlockPatcherSampler](https://github.com/cubiq/Block_Patcher_ComfyUI)
  * Warning, this model has discovered NSFW and can randomly produce a NSFW image when not expecting one.
  * 58 block layers
  * 12 runs per single block layer to find best weights 0.3 to 1.4
  * 696 total weights tested block by block to get 58 weights for the merge 
  * So something like this, pick the best pic from each block layer
  * ![](version_3e_images/m1.png)
  * Get all the block weights hand picked, merge the model with the new weights.  Isn't there a lot of variables to this?  Sure, but I kept it close to my normal gens and after testing, it seems fine with other samplers, schedulers, loras, etc..
  * ![](version_3e_images/m2.png)
  * For Science:
  * ![](version_3e_images/standard_v3_nongguf_3.png)
  * ![](version_3e_images/standard_v3_nongguf_2.png)

[Civitai Gallery of Schnell Images from this model](https://civitai.com/collections/1216945)


# EclecticEuphoria_Flux_Unet_Schnell_v2

A much improved version with a constant ability to reproduce better consistant images.  Q6 and Q4 models produce much better images out of the gate now.  Does NSFW, but with the typical flux anatomy issues, if you struggle with flux not wanting to do NSFW try adding "exposed (breast:2.0), (Nipples:2.0)"  You will need the Vae and Dual Clip as these are unet models.

![](version_2_images_wf/v2.png)
![](version_2_images_wf/Flux_Schnell_Multi_Cast_Loader_WF_v2.png)

[EclecticEuphoria_Flux_Unet_Schnell_v2-f16.safetensors](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_f16_v2.safetensors?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v2-Q8_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v2-Q8_K.gguf?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v2-Q6_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v2-Q6_K.gguf?download=true)

[EclecticEuphoria_Flux_Unet_Schnell_v2-Q4_K.gguf](https://huggingface.co/ND911/EclecticEuphoria_Flux_Unet_Schnell_models/resolve/main/EclecticEuphoria_Flux_Unet_Schnell_v2-Q4_K.gguf?download=true)

## GGUF Q8 - refinded and upscaled
![](version_2_images_wf/20240917110328_Q8_GGUF_v2.png)

## GGUF Q6 - refinded and upscaled
![](version_2_images_wf/20240917110328_Q6_GGUF_v2.png)

## GGUF Q4 - refinded and upscaled
![](version_2_images_wf/20240917110328_Q4_GGUF_v2.png)

## Link below are more images - the latest, most upper ones
[Civitai Gallery of Schnell Images from this model](https://civitai.com/collections/1216945)

-------------------------------------------------------------------

# EclecticEuphoria_Flux_Unet_Schnell_models (Initial Release)

Not of a huge fan of Schnell on first use so I took up a challenge to remove what I didn't like of Schnell.  First goal was to remove the hideous Greenish blotchy look and the random dark splotches all over.  I used lora and thanks to the negative field was able to remove most of those issues and get a more natural looking image.  So I am calling this version a win.  The standard model produces much better images out of the gate vs the ggufs.  A refiner is definately needed in both cases, the ggufs are complete rubish looking straight out of the sampler, but it's a good thing as the noise gives the refiner something good to work with.  So find a favorite refiner model and plug it in for best results.  Images are created in 8 steps.  I only tested the 8_0K gguf a couple times, not sure how well the K6 and K4 models do.  I do like the artsy style of schnell tho, it can really produce some unique beatuiful pictres in a hurry.

## Standard Workflow

![](Flux_Unet_Schnell_WF.png)

## GGUF Workflow
![](Flux_Schnell_GGUF_WF.png)

## Standard Comparison sampler vs refine

![](sampler_vs_refined.png)

## Standard

![](20240915040224_USDBF_END_Standard_.png)

## GGUF Q8

![](20240915164748_USDBF_END_GGUF_.png)

## Standard with controlnet

![](20240915_Schnell_with_controlnet_.png)


## Link below are more images - the latest, most upper ones
[Civitai Gallery of Schnell Images from this model](https://civitai.com/collections/1216945)