You are on page 1of 10

Table of Contents

1. Disclaimer..............................................................................................................................................3
2. Multidiffusion upscaler..........................................................................................................................3
3. Region prompt control...........................................................................................................................3
3.1 Settings............................................................................................................................................3
3.2 Workflow.........................................................................................................................................5
3.2.1 Model, prompt, settings etc….................................................................................................5
3.2.2 First region...............................................................................................................................6
3.2.3 Second region..........................................................................................................................7
3.2.3 Third region.............................................................................................................................8
3.2.3 Fourth region...........................................................................................................................9
3.3 Result.............................................................................................................................................10
4. The end.................................................................................................................................................10
1. Disclaimer

This is mainly tutorial on personal workflow on how i use multidiffusion upscaler for automatic1111.
I might get something wrong and if you spot something wrong with tutorial, please leave a comment.
Any feedback is welcome. Tutorial page https://civitai.com/models/34726

I am not the creator of this extension and i am not in any way related to them. They can be found from
Github page below. Please show some love for them if you have time :).

https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111

2. Multidiffusion upscaler

In this tutorial I will be using Multidiffusion upscaler. If you have not read the first part of the tutorial:
“1.How to use...” and you have not used Multidiffusion upscaler before, I recommend reading it first.
Second part of the tutorials gave more information on img2img upscaling.
This is third part of the tutorial and will be focusing on Region Prompt Control of multidiffusion.

3. Region prompt control

With region prompt control you can make regions. Every region has own prompt for positives and
negatives. You can decide the area in the image which it affects.

3.1 Settings

Settings are pretty simple


and straight forward. With
enable we enable the Region
prompt control. You have to enable Tiled Diffusion for it to work! Draw full canvas background: if
you don't use background in regions and want to use foreground to add object to your image use this. If
you have background in the regions, leave this disabled.
Pressing “Create txt2img canvas” creates area for us
where we can see regions when we enable them.
Every time you change your image
size you have to create canvas again!
In this image I have 4 regions enabled.
You can move and resize the regions
from the canvas or from the region
settings.

Region settings look like this:

You have to enable every region you


are using. When you enable the region
it appears in the canvas as colored
area.

Type: There are 2 different kind of


types for region
• Background: It is good idea to
fill entire canvas with at least 1
background.

• Foreground: With this you add


things to your background.
When you have foreground
enabled new setting called
“Feather” will appear.

With this setting you decide


how much you want to blend
the region to your background.

With 0 it does not blend it at all and creates


completely “own picture” and at 100 it will
give it max blend and most likely does not
add anything to the image.
3.2 Workflow

In this workflow I will go through simple workflow that shows example of how to use regions and how
it affects the image step by step. I recommend experimenting after you learn how to use it.

3.2.1 Model, prompt, settings etc…

Model: revAnimated:
https://civitai.com/models/7371/rev-animated
VAE: vae-ft-mse-840000-ema-pruned
Main prompt is used for quality, camera, lighting
tags.

I will be using settings like these for the tutorial:

I am using high Width and Height to get more detail


from the regions.

You can use Hires. Fix to scale the image. I will be


using img2img upscaling with tiled diffusionf or this
method. You can read the second tutorial “2
img2img workflow” to get general idea how it
works.

The seed in here does not affect the image, unless


you have “Draw full canvas” enabled

In below you can see my tiled diffusion settings.


What these do are explained in the first tutorial. “1.
How to use+T2IWorkflow”
3.2.2 First region

First I will be adding Background region. To


keep it simple I am going to add positive
prompt “Field with flowers”. Testing it with
generate gave me girls, so I dropped them into
negatives. (~ ̄▽ ̄)~

I generated it few times, and found decent


looking field and locked the seed. You can
either do these 1 by 1 with generating
between adding each region and finding seeds
you like, or you can just add every region and
then start generating. Both work.

I find it fun to add each region one by one and


see how the image builds.

First region covers


the whole canvas.

This is the image with the first region


enabled. It is just a field with flowers, just
what I wanted. :)
3.2.3 Second region

For every region after 1st Type will be


foreground.

What is missing? Yeah water. Lets add a river.


I could use river as prompt, but I decided to
go with flowing water. Feather 0.25 sounds
good. Usually something between 0.20-0.40 is
good blend.
This was generated few times to find nice
image, after I found it, I locked seed.

Lets add it in the middle like this.


(Now that I look at it. Its
not in the middle at
all… Cant go changing
it now that I have
generated everything ;()

The generated image is not exactly what I had


in mind, but I am pleasantly surprised.

Still missing something 🤔


3.2.3 Third region

I know. Lets add a house. Living in this kind


of place would be nice. No other people
bothering you, just chilling. (~ ̄▽ ̄)~

Lets add nice small wooden hut with 0.25


feather. Sounds nice.

Lets drop it next to the river on the right side


The yellow box in the
canvas.

This worked better than I expected. It did not


mess the image and I got nice hut to live in
now.

Still could add something there 🤔


3.2.3 Fourth region

Lets add a nice deer on the other side of river.


Sounds like a plan.
I am going to use higher feather for this 0.45
could work. I only want the deer and nothing
else.

As I don’t want only deer head, which I first


got with only “deer” in the prompt, I am
going to use “deer full body”. This hopefully
gives me a full deer.

The deer is the small


green square in the
canvas.
If you are using lower
resolution for your
image, you might need
bigger squares.

I got full deer (~ ̄▽ ̄)~. Not just head.

I think the image looks perfect and this should


convey the general idea of the region control.
If you have any questions please leave them
in the tutorial page.
3.3 Result
The final image can be seen on the first page. The image was scaled in img2img with tiled diffusion +
tile resample from controlnet. The next tutorial might be for controlnet + tiled diffusion workflow or
inpaint.

4. The end
I recommend testing and experimenting with different kind of settings to master the region prompt
control.
Thank you for getting this far in this tutorial. English is not my first language so forgive me for
mistakes I make. You can point them out in the discussion of the tutorial in civitai website ;)

https://civitai.com/models/34726

Big thanks to pkuliyi2015 https://github.com/pkuliyi2015 for creating this extension

https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111 if you like the extension,


please go and give some love.

Please leave feedback, review, and images you have generated! Those will motivate me to do
more of these guides. 〜( ̄▽ ̄〜) (~ ̄▽ ̄)~

Potatovision https://civitai.com/user/Potatovision/models

You might also like