Your noob to expert guide in making great AI art.

[BlueLover69-38]

Thanks for replying,
no i don't have it, my image is fully generated i believe,
here is my full interface


Edit:
i figured out just now, all i need to do is to just press the blue check icon in the bottom,
thank you so much for this tutorial you are a awsome

 

[tatiana.kaluzny]

NoTraceOfLuck

Very good guide! Took me from zero knowlege to toying around for a couple of hours.

I've also went and read the thread for your game. It's really interesting to see you develop over time and find different solutions to your problems.

One thing that I'm not sure how you do is how to get consistent element over several images. You obvious make a lot of images and your characters always have the same face, shape, etc. Obviously you're referring to some well-known characters so that makes it easier. But do you use one of the tools/techniques taught here to achieve that or there's more?

 

[NoTraceOfLuck]

NoTraceOfLuck

Very good guide! Took me from zero knowlege to toying around for a couple of hours.

I've also went and read the thread for your game. It's really interesting to see you develop over time and find different solutions to your problems.

One thing that I'm not sure how you do is how to get consistent element over several images. You obvious make a lot of images and your characters always have the same face, shape, etc. Obviously you're referring to some well-known characters so that makes it easier. But do you use one of the tools/techniques taught here to achieve that or there's more?

The secret is my characters are actually not as consistent as they seem

To keep characters appearing consistent, I do a list of different things:

1. Generate a lot of images. For every 1 image in my game, I have probably generated 20-50 other versions of that image. You will never get consistency if you just take the 1st result every time.
2. You can manually fix the body and face shape using the techniques I showed in my 'ControlNet' tutorial. For example, you can literally just erase parts of the image and draw new white lines in the shape you want. You don't need to be a great artist to draw slightly large or smaller breast outlines, or give slightly better curves to the character:

​

3. The eyes are very important. I use one specific model to replace the eyes on every character. The first thing people will notice is the eyes, and if you keep those consistent, the whole image feels much more consistent. To do this, I just follow the technique I posted for "Inpainting"

4. Finally, I use some "movie magic" in my game. I put a very short fade-to-black after every time I switch background images. This makes inconsistencies between the 2 images much harder to notice.

 

[tatiana.kaluzny]

3. The eyes are very important. I use one specific model to replace the eyes on every character. The first thing people will notice is the eyes, and if you keep those consistent, the whole image feels much more consistent. To do this, I just follow the technique I posted for "Inpainting"

That's actually very smart. Thank you.

Life is a bit much right now, but I will definitely keep toying with this and will follow your game!

 

[NoTraceOfLuck]

That's actually very smart. Thank you.

Life is a bit much right now, but I will definitely keep toying with this and will follow your game!

Good luck, I look forward to seeing what kind of things you create

 

[not_a_user_69]

NoTraceOfLuck Sorry for tagging you.
I just wanted to ask about the GPU.
So, I am planning to upgrade my GPU, I have 3 GPUs in mind, I can go for 5060ti 16gb (within my budget and thinking of buying Asus), then 4070 12gb(but over budget), then 5070 12gb(very over budget need to change PSU too).

So for invoke AI, is 12gb sufficient or is 16gb needed?
Should I go with 4070 or 5060ti?
If not these, 5070?
Please enlighten me here, I am damn confused.
People who have ideas can also help me.
I am using WAI-NSFW-illustrious

 

[n3k0d3v]

NoTraceOfLuck Sorry for tagging you.
I just wanted to ask about the GPU.
So, I am planning to upgrade my GPU, I have 3 GPUs in mind, I can go for 5060ti 16gb (within my budget and thinking of buying Asus), then 4070 12gb(but over budget), then 5070 12gb(very over budget need to change PSU too).

So for invoke AI, is 12gb sufficient or is 16gb needed?
Should I go with 4070 or 5060ti?
If not these, 5070?
Please enlighten me here, I am damn confused.
People who have ideas can also help me.
I am using WAI-NSFW-illustrious

It's simple really, always go for more VRAM and 16GB should be the bare minimum nowadays. I have the 5060 Ti 16GB and am quite satisfied with it. Or you could wait for Nvidia to release the newer 50xx GPUs which are supposed to have more VRAM, though if you can not wait then yeah get the 5060 Ti which should be within your budget.

You will be able to squeeze out more performance with this Blackwell GPU generation(50xx) because it supports FP4, which is very good for Nunchaku quants for example, so try to avoid older Nvidia GPU generations.

I started with InvokeAI but had to also start learning ComfyUI(shiny new tools, like Qwen-Image Edit and others) and i think it was a good decision to get 5060 Ti.

 

[NoTraceOfLuck]

NoTraceOfLuck Sorry for tagging you.
I just wanted to ask about the GPU.
So, I am planning to upgrade my GPU, I have 3 GPUs in mind, I can go for 5060ti 16gb (within my budget and thinking of buying Asus), then 4070 12gb(but over budget), then 5070 12gb(very over budget need to change PSU too).

So for invoke AI, is 12gb sufficient or is 16gb needed?
Should I go with 4070 or 5060ti?
If not these, 5070?
Please enlighten me here, I am damn confused.
People who have ideas can also help me.
I am using WAI-NSFW-illustrious

I agree, 5060ti is your best option. It will be a little bit slower for each image, but will allow you to do higher resolution images and makes video generation a possibility.

 

[Jansonpierre]

NoTraceOfLuck Sorry for tagging you.
I just wanted to ask about the GPU.
So, I am planning to upgrade my GPU, I have 3 GPUs in mind, I can go for 5060ti 16gb (within my budget and thinking of buying Asus), then 4070 12gb(but over budget), then 5070 12gb(very over budget need to change PSU too).

So for invoke AI, is 12gb sufficient or is 16gb needed?
Should I go with 4070 or 5060ti?
If not these, 5070?
Please enlighten me here, I am damn confused.
People who have ideas can also help me.
I am using WAI-NSFW-illustrious

I have a 5060ti and it works just fine, 4070 is pretty good but if its over ur budget, 5060ti 16gb is perfect. I got a Ryzen 7 7800X3D + RTX 5060 Ti 16GB.

 

[Jansonpierre]

I agree, 5060ti is your best option. It will be a little bit slower for each image, but will allow you to do higher resolution images and makes video generation a possibility.

btw May I ask what content you are working on for the next update?

 

[NoTraceOfLuck]

btw May I ask what content you are working on for the next update?

You mean for this guide? Or for my game?

My next update for this guide will probably be about Flux Kontext / image editing models, but I am waiting for Invoke to add support for Qwen Image Edit hopefully before I do that.

For my game I posted some small updates on the game thread.

 

[Madhealzz]

Great guide! Thank you for the information! I noticed that you didn't mention upscaling.

For your game I assume you generate at a 9:16 aspect ratio (to match Renpy's resolution) and then upscale in invoke? or do you have a separate workflow for that?

 

[not_a_user_69]

Great guide! Thank you for the information! I noticed that you didn't mention upscaling.

For your game I assume you generate at a 9:16 aspect ratio (to match Renpy's resolution) and then upscale in invoke? or do you have a separate workflow for that?

Yes, I was wondering about this too.
If I use dimension which is not supported by models (Default), it sometimes gives weird output, legs look stupid.
Please provide complete guide to that from Invoke, upscaler option.

 

[NoTraceOfLuck]

Yes, I was wondering about this too.
If I use dimension which is not supported by models (Default), it sometimes gives weird output, legs look stupid.
Please provide complete guide to that from Invoke, upscaler option.

Great guide! Thank you for the information! I noticed that you didn't mention upscaling.

For your game I assume you generate at a 9:16 aspect ratio (to match Renpy's resolution) and then upscale in invoke? or do you have a separate workflow for that?

I can add a full section on upscaling at some point, but the general idea is that there are 2 ways to do upscaling.

Option 1: The upscale tab

You can simply right click and image and click 'Send To Upscale' and it will bring you to the upscale tab. This tab is very simple, you just select a model and a resolution multiplier. The 'best' model will depend on the type of image. Realistic images have different models than anime images, so you'll want to test them out.

Option 2: Upscale with ControlNet

This is what I use for images in my game, and gives more control than the previous option if there are specific details you want to retain or a specific resolution you want to hit.

Steps:

0. Generate an image using a supported resolution that does not produce weird legs, etc...

1. Send the image to the canvas (right-click an image -> new canvas from image -> as raster layer)

2. Adjust the width and height of the canvas to your desired upscaled resolution. You will notice that the bounding box grows larger than your actual image and will contain a bunch of 'empty' space. That's normal.



3. Right click the raster layer that was created, and click 'transform', then drag your image to fill the entire bounding box. Then click Apply. If the image is not exactly the same aspect ratio, it may need to be cropped.




4. To crop the image, right-click the canvas and select 'Crop Canvas to BBox'



5. Now create a controlnet of the raster layer. (Right-click the raster layer -> Copy Raster Layer To -> New Control Layer)



6. On the new control layer, select Canny model

7. Now, you have 2 options here. You can either fully disable the original raster layer, and have the model generate a new, fresh image that is not influenced by the original image. This can be good if you want to redo the lighting or change some part of the image. OR you can keep the original image. This is good if you're 100% happy with the original image, but just want it to be a higher resolution.

8. If you want to keep the original image, you can adjust the denoise slider, then simply regenerate the image. To get the best quality, you'll want a denoise of at least 0.65. The higher you go, the more detail of the original image that will be lost. I usually stick with around 0.65 to 0.8

9. If you don't want to keep the original image, disable the raster layer by click the circle on the right side of the layer (next to the trash can). Now simply regenerate the image.


-- -- -- -- --Old (low res) -- -- -- -- -- -- -- -- -- -- -- -- -- --- -- -- -- -- -- New (high res)


Doesn't look too much different but you'll notice higher detail in the eyes especially

 

[Madhealzz]

Thank you for such a detailed response! Much appreciated!

 

[not_a_user_69]

NoTraceOfLuck So, I’ve run into an issue and I’m not sure how to deal with it.
The problem occurs when I generate an image at 1368(w) × 768(h) — I end up getting double characters (duplicates).
How can I solve this issue? When I use the default resolution or a mobile-type resolution, it doesn’t happen.
This issue only occurs when I use a desktop-type resolution.
Same female character cloned right beside it.

 

[osanaiko]

The image gen models like to fill the space. Sometimes you can avoid it via prompting, especially negative prompts, but not always.

There's a few ways to encourage the result you want:

1. use regional prompting extensions

2. Try generating the character using a portrait size ratio of dimensions. Then use resizing, manual editing to copy the character image on top of a pre-generated background, then use img2img to blend the styles/lighting/seams.

3. Or also you can try generating a single character in portrait ratio and then use and img2img "outpainting" to extend to the size you want.

4. use img2img with stick-figure / color region simple drawings as source images to get the layout you want

 

[NoTraceOfLuck]

NoTraceOfLuck So, I’ve run into an issue and I’m not sure how to deal with it.
The problem occurs when I generate an image at 1368(w) × 768(h) — I end up getting double characters (duplicates).
How can I solve this issue? When I use the default resolution or a mobile-type resolution, it doesn’t happen.
This issue only occurs when I use a desktop-type resolution.
Same female character cloned right beside it.

In addition to what osanaiko said, this can be an issue with prompting.

If you prompt something that generally involves 2 people, it will often clone your character and put 2 of them in. For example, if you having "kissing" in your prompt. it will draw your character kissing themselves. It can also happen if you have conflicting terms in your prompt. For example, if you have "Front view" and "Side view" in the same prompt, it may try to draw your character both at a front view and side view, and the only way it can do that is to duplicate your character.




It can also be an issue with the model or any LoRAs you're using. Some models were trained on images of multiple people and so any image you generate will be affected by that. I've also found that having VERY long prompts tends to make the problem worse.

When I encounter this issue, I usually just do the very simple solution: inpaint out the duplicate manually.

----->

 

[Madhealzz]

The problem could be as simple as not using "1girl" tag, or whatever the equivalent for your model is

 

[not_a_user_69]

NoTraceOfLuck I know I am troubling you, here is the Positive prompt
Detailed hair, detailed face, detailed body, detailed background, beautiful eyes, score_9, score_8_up, score_7_up, masterpiece, best quality, amazing quality, highly detailed, (best shadow), (absurdres), HDR, highres, absurdres, detailed face, detailed eye, face focus, (mature woman:1.2), 1girl, (full shot:1.5), hero view, (front view:1.4), long black hair, emerald eyes, seductive gaze, makeup, lipstick, sweat, medium breasts, milf, sexy, detailed skin, photorealistic lighting, female teacher attire, sexy teacher, office lady, open shirt, (shirt unbuttoned:1.2), cleavage, visible collarbone, tight skirt, blazer open, sitting on desk, (dominant pose:1.1), seductive expression, confident, professional attire, classroom, blackboard, indoor, school setting,

Please craete a tutorial by generating an image for 16:9 ratio and upscaling it.
If I create using default resolution or 9:16, upscaling it to 16:9 FHD doesn't give satisfactory output.
and when I create using 16:9 resolution character is created side by side losing alignment, is there anything which could be done?
Simpler steps cause to create multiple images for 16:9 i need a efficient method. I know I am troubling, but please help me here.

Or is there any Lora available in CivitAI to tackle it?