File size: 1,525 Bytes
27817a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
48ea2a7
27817a6
48ea2a7
27817a6
48ea2a7
 
 
 
f97c939
 
 
 
48ea2a7
 
27817a6
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
---
license: apache-2.0
---

# EclecticEuphoria_Flux_Unet_Schnell_models

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|Yellowish look and the random daark 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)