Quick Preface

Some caveats or thoughts;

this is written with the idea you are comfortable with text-to-image prompting and
you currently have a Text-to-Image workflow set up
This is focused with ComfyUI in mind.
This is focused on 'Clip' | Text Encoding a latent image

If not, drop a comment - I'll see what I can do to help get you started.

The guide is more to start with an initial prompt and then subtract / add more to it.

What I've wanted to do the most is manipulate an image and replicate what I see

Its incredibly varied, the prompts order of words (or order of operations as I see it) does seem to matter before we get into altering the model and sample. The syntax and grouping of the prompt / caption does matter. Even adding a ",," matters, which you will see in the sample prompt.

Of course this is not all encompassing and very very long, if anything, this is to illustrate the variety of image when all else is the same by simply adding or removing a caption as part of your prompt.

Setup

Before that, lets start with some settings for consistency.

Model: SDXL | epiCRealism XL

https://civitai.com/models/277058?modelVersionId=1074830

Start with a latent image resolution 744x1152

Sampling settings:

Bear in mind of Embeddings - like bad hands

( I tried to track it down but dont recall where I found it to give credit | its a PT file -> attached )

Prompting

Basics

1 word or two word caption
- Format can be: (these do give different results)
  - with a space
    - gothic room
  - with an underscore
    - gothic_room
separated by a comma ,

not a catch-all, jut what I have observed

Advance

Multiple adjectives as one prompt -
different syntax, same adjectives, different image
- <very detailed eyes,detailed lips,detailed face,cute,pretty>
- (very detailed eyes,detailed lips,detailed face,cute,pretty)

Lets Start Prompting - In Detail

Full prompt (Clip) of the original image

Positive

portrait, close up, cute face, beautiful,18 years old, (black bold eyeliner, makeup:1.3), pale skin, young,bangs,freckles, ear,wavy brunette hair,freckles,,interior,fancy bedroom,gothic room,castle interior,petite body, small breasts, beautiful legs, realistic soft skin, looking at viewer, hyperdetailed, analog photo, Cinematic,contrast,high contrast,(very detailed eyes,detailed lips,detailed face,cute,pretty)

Negative

bad hands, bad anatomy, ugly, deformed, (face asymmetry, eyes asymmetry, deformed eyes, deformed mouth, open mouth)

Starting image

Removing portrait,

Removing close up,

swapping 18 years old => 1girl (I mean whatever that translates to | moving on)swapping 18 years old => 1woman

Removing cute face,

Removing beautiful, ( Did I really subtract much by removing cute face, beautiful? )

Although, I'm not sure why her hand is by her face.

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

We should be at:

1 woman, (black bold eyeliner, makeup:1.3), pale skin, young,bangs,freckles, ear,wavy brunette hair,freckles,,interior,fancy bedroom,gothic room,castle interior,petite body, small breasts, beautiful legs, realistic soft skin, looking at viewer, hyperdetailed, analog photo, Cinematic,contrast,high contrast,(very detailed eyes,detailed lips,detailed face,cute,pretty)

Removing analog photo,

Removing Cinematic,

Adding analog photo,

Swapping out analog photo for Cinematic (Not interesting other than you should get the same photo back as above, showing off if only that is changed, you still get the same image)

Lets remove Cinematic, and

contrast, high contrast,

Lets add high contrast, back | Now I dont see a difference between keeping one or the other contrast prompts

removed high contrast, and added contrast back

remove high contrast, you will be back to one of the above images

Lets remove hyperdetailed,

I guess at this point we get to see more and more of the model, definitely not intentional

1 woman, (black bold eyeliner, makeup:1.3), pale skin, young,bangs,freckles, ear,wavy brunette hair,,freckles,interior,fancy bedroom,gothic room,castle interior,petite body, small breasts, beautiful legs, realistic soft skin, looking at viewer, (very detailed eyes,detailed lips,detailed face,cute,pretty)

We're here (same neg) (above image)

To move closer:

add close up,

or portrait ( adding choker, necklace, or 'detailing' the jewelry will touch up her necklace)

Lets fix that necklace

((

1 woman, (black bold eyeliner, makeup:1.3), pale skin, young,bangs,freckles, ear,wavy brunette hair, necklace,freckles,interior,fancy bedroom,gothic room,castle interior,petite body, small breasts, beautiful legs, realistic soft skin, looking at viewer, (very detailed eyes,detailed lips,detailed face,cute,pretty)

))

remove necklace

adding

close up, portrait, (oh look a crown has been added - changing the seed will remove the crown)

adding photo-realistic,

removing close up, portrait,

Adding necklace back

(ok she looks familiar)

Removing necklace, photo-realistic

================

Lets change the Venue

Removed interior, fancy bedroom,gothic room, castle interior,

adding Beach, ocean, palm-trees,

( and we are zooming out to get the new prompts in to the image )

I guess we can add gothic-dress

eh, I didnt like gothic-dress

Lets try gothic dress,

add a beach chair, (I guess its confused with the order of operations)

Beach, ocean, palm-trees, gothic dress, beach-chair,

Beach, ocean, palm-trees, beach-chair, gothic dress,

===============

Fun with syntax... (bolded and italicized )

Original ()

1 woman, (black bold eyeliner, makeup:1.3), pale skin, young,bangs,freckles, ear,wavy brunette hair, ,freckles,

interior, fancy bedroom,gothic room, castle interior,

petite body, small breasts, beautiful legs, realistic soft skin, looking at viewer, (very detailed eyes,detailed lips,detailed face,cute,pretty)

Changed () to <>

1 woman, (black bold eyeliner, makeup:1.3), pale skin, young,bangs,freckles, ear,wavy brunette hair, ,freckles,

interior, fancy bedroom,gothic room, castle interior,

petite body, small breasts, beautiful legs, realistic soft skin, looking at viewer, <very detailed eyes,detailed lips,detailed face,cute,pretty>

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

Assuming you made no other alterations to your Model, sampler, latent image, etc

This prompt will get you back to the starting image

================================

Bonus:

I find incredible variance from different samplers and 'scheduler'.

This is Left Image feeding into Right

I don't dare to re-tread sampler explanation, but if you ever were curious or want a refresher

https://stable-diffusion-art.com/samplers/#Samplers_overview

Fun with Prompts