Example way to make SF2, other Sprite faster by AI - good quality

Started by alongcaca, November 28, 2025, 02:46:27 PM

Previous topic - Next topic

alongcaca

i use nano banana pro service , 2 way
Way 1
Source


Result

Prompt
Recreate the character on the right by keeping all the original colors, skin texture, muscle shine, highlight-shadow, lighting and shading style of the character on the right.

Keep all the details like beard, face, gold belt, gold wrist, pants, shoes, belt and all accessories of the character on the right.

However, change the pose of the character to be exactly like the standing pose of the character on the left: two legs standing wide, two hands on the hips, standing straight and symmetrical.

The pixel-art style must be kept the same as the original image, no new details added


Way 2:
Source

Result

Prompt:
Recreate the character on the left by copying all the colors, skin texture, muscle shading, lighting, shading, textures, and accessories from the character on the right. Keep the pose, stance, body proportions, and layout of the character on the left exactly the same.

The new character must have the same brown skin color, muscle shading, highlights, shadows, and shading style as the character on the right.

All accessories such as the gold belt, gold wristband, shoes, pants, belt, etc. copy the style and color of the character on the right but place them in the correct position of the character on the left.

Do not change the pose, stance, or basic muscle shape of the character on the left.

Keep the classic pixel-art style true to the original

Make sprite sheet easier


     Posted: November 28, 2025, 03:05:52 PM
Bonus
change ken sprite like left man

Noside


alongcaca

I tried the whole sprite sheet but it didn't seem to be as expected, so dividing it into smaller parts would be better for image recognition, it seems that in the future or now just need to adjust the prompt better then the result will be even better

Currently according to my prompt, it needs a little editing to be able to put it into a character

Here are the sources I've made,
Reference image

Sprite image

Result 1

Result 2

Chat frame image

prompt
An image with one character is a reference image; an image with one character in multiple poses is the final result.

Recreate the reference character by keeping all the colors, skin texture, muscle gloss, highlight-shadow effects, lighting style and shading of the character on the right.

Keep all the details like beard, face, gold belt, gold wrist, pants, shoes, belt and all accessories of the reference character.

Copy and replace all the poses of the resulting image to the character of the reference image, for all the images in the image

The pixel-art style must be kept the same as the original image, no new details added.

Final result: Keep the pixel size of the character on the right as the original image, dark blue background

Hopefully it will help you have another way to create characters faster

Funk


FeLo_Llop

- Why do I like AI?- Because it can help the spriter in a overheat situation or when don't know how to do a particular sprite. The AI can help making the intermediate frame. All that's needed is to overtrace/ draw above that frame(using a layer, for example) while using the required colours (AI blurs them a bit).
- Why don't I like AI? Because even helpful, it can't swap the spriter genious.

My advice, Alongcaca, is you to learn and learn to improve your nowadays work; using AI? yes, but don't expect it to do the whole thing for you.

Practice!! You'll one day become a master!

[ Rose/ C.Viper/ Mars People/ Ash Crimson / Ivan Ooze ]: Finished
[ Shiki / Mature ] W.i.P.
[ ?? / ?? ]: Possible W.i.P.

Basara Bogard

I like what you're trying to do, and there's nothing wrong about using AI as a tool instead to let it make all the job.

Good job trying to recreate Darun Master sprites with this method, but should be better if you use Zangief sprites for his moves instead of Ken's. Also, not sure if you're making the sprites for a char or as a future resource for other creators. Keep it up!!

alongcaca

After a few days, I tried germini 3 pro again, I realized that when you put a new sprite in and ask it to redraw, it will expand its pose library, structure into its database, so the more images it doesn't have, the more accurate the results it will redraw later.

Ref image

Source pose

Result

I know there are many different opinions on using AI to create the results, thanks for your feedback
I have a question because the current AI cannot make a standard image background, so from this to a beautiful sprite there are still many steps, does anyone know a guide to quickly and beautifully separate the background?

tunglashor

As to how useful this will be for creating characters, I'll defer to the spriters who know what they're talking about, but I just wanted to say I find this super cool, thanks for sharing it  :8):

When you mention a "standard image background" I'm not sure what you mean, is it that the image isn't generated with an indexed palette?

alongcaca

Quote from: tunglashor on December 06, 2025, 12:25:49 AM
As to how useful this will be for creating characters, I'll defer to the spriters who know what they're talking about, but I just wanted to say I find this super cool, thanks for sharing it  :8):

When you mention a "standard image background" I'm not sure what you mean, is it that the image isn't generated with an indexed palette?

Yes, it has a lot of noise and not  indexed palette

PotS

You can help with Ikemen GO's development by trying out the latest development build and reporting any bugs on GitHub.
My Mugen and Ikemen content can also be found here.

MGMURROW

Not exactly sure how this works but could it be used to complete the SF3 Bison and SF3 Strider sprite sheets? Or how about making shock animations for the characters who don't have them? maybe the falling twist animations from SF3? And the Jedah grab move where he pumps blood into P2 or Dimitri's move where he turns the characters gender?

Just a thought

MGMURROW

alongcaca

Quote from: MGMURROW on December 11, 2025, 11:44:28 AM
Not exactly sure how this works but could it be used to complete the SF3 Bison and SF3 Strider sprite sheets? Or how about making shock animations for the characters who don't have them? maybe the falling twist animations from SF3? And the Jedah grab move where he pumps blood into P2 or Dimitri's move where he turns the characters gender?

Just a thought

MGMURROW

i think AI possible to help you complete SF3 Bison,

My method involved using two sprites: one as a reference, and the other containing many necessary poses. The AI ​​would automatically convert them using the correct prompt (I've included an example prompt above).

The next step is to take the sprite you want to convert and place it within a single large image. Then, practice with multiple prompts until you achieve the desired result.

alongcaca

Update of gemini, nano banana make full sprite sheet easier
Now see how it improve,  i have 2 source of image ( i try to make cvs2 style )
1 is front and back of new char

2 is some pose for copy

3 here is result

Here is the prompt
Two input images:

Image A = character appearance reference (front and back view of the same character).
Image B = sprite sheet reference containing multiple poses and animation frames.

Task:
Replace the character in Image B with the character from Image A.

Use Image A strictly as the visual appearance source.
Use the front view of Image A for front-facing poses.
Use the back view of Image A for rear-facing poses.

Keep EXACTLY all appearance details from Image A:
– skin color and tone
– muscle shading and gloss
– highlight and shadow placement
– facial features, beard, hair
– crown, skull garland, belt, wristbands
– cloth, ornaments, accessories
– exact pixel color palette consistency

Preserve EXACTLY all poses, anatomy, proportions, animation timing, spacing, and frame layout from Image B.

------------------------------------------------------------
MASTER SIZE NORMALIZATION RULE
------------------------------------------------------------

Choose one specific sprite frame from Image B as the master size reference.

Use that sprite's exact pixel height (from feet to top of head) as the absolute height standard.

All other frames must be resized proportionally to match this exact pixel height.

All frames must fit inside the exact same bounding box dimensions as the master sprite.

Maintain consistent sprite alignment and anchor position.

Do NOT:
– crop
– stretch horizontally
– stretch vertically
– distort anatomy
– alter proportions
– change frame spacing
– modify camera angle

Maintain strict pixel grid integrity.
Nearest-neighbor scaling only.
No interpolation.
No subpixel rendering.

------------------------------------------------------------
RENDERING STYLE REQUIREMENT – CAPCOM VS SNK 2
------------------------------------------------------------

All recolored sprites must follow the exact shading and coloring logic of Capcom vs SNK 2 arcade sprites.

– strong contrast cel shading
– 2–3 tone shading per color region
– sharp hard-edge shadows
– no gradient blending
– no soft transitions
– high saturation arcade palette
– bold shadow blocks
– clear muscle volume highlights
– metallic gold highlights rendered as pixel clusters
– deep contrast under muscles and cloth folds
– SNK-style highlight placement on limbs and torso
– no modern digital painting
– no semi-realism
– no painterly shading
– no lighting reinterpretation

Color treatment must strictly resemble early 2000s SNK / Capcom arcade sprite rendering.

Do not modernize the shading.
Do not smooth highlights.
Do not add bloom or lighting effects.

------------------------------------------------------------
STYLE CONSTRAINTS
------------------------------------------------------------

– strict pixel art
– original pixel resolution
– hard pixel edges
– no anti-aliasing
– no smoothing
– no repainting
– no soft shading
– no gradient blending
– no added lighting
– no reinterpretation
– no redesign
– no extra detail
– no stylistic enhancement

------------------------------------------------------------
OUTPUT REQUIREMENT
------------------------------------------------------------

Output a complete sprite sheet with all poses from Image B,
fully reskinned using the character design from Image A,
with uniform pixel scale across all frames.

Exact pixel dimension consistency required across the entire sheet.

Background:
transparent color only.
no lighting on background.
no blur.

Momotaro

Interesting, but these sprites are far for usable in a game.
All colors are messed up this is not pixel art to me.

snowy997

It doesn't look bad, it just looks blurred, i think you can use ai to do some sprites with the right prompts

alongcaca

Quote from: Momotaro on February 27, 2026, 11:25:31 PM
Interesting, but these sprites are far for usable in a game.
All colors are messed up this is not pixel art to me.
Actually, even after getting this result, some manual steps are still needed to make the spiky usable. However, it's still faster than doing it 100% manually.

Quote from: snowy997 on February 28, 2026, 12:51:51 AM
It doesn't look bad, it just looks blurred, i think you can use ai to do some sprites with the right prompts
Yes, with further research, it might be feasible.





PotS

Simply applying the indexed palette to the result will probably go a long way.
You can help with Ikemen GO's development by trying out the latest development build and reporting any bugs on GitHub.
My Mugen and Ikemen content can also be found here.

alongcaca

While continuing my search for a better way to create sprites, I stumbled upon another AI tool that helps generate each sprite individually, as per the reference. However, this approach is more complex than the previous one because each sprite must be created separately. Please see the content below.
Here is the 1st

Prompt
The man turned to the side in a 2D fighting game perspective, then performed a somersault kick backward.

Here is the link of this tool
https://huggingface.co/spaces/selfit-camera/Omni-Image-Editor

I think this tool is only suitable for HD characters; it's not suitable for pixel-quality characters.

DatKofGuy

Would this method work for trying to replicate character portraits? Like say KOF2002 or 2003 winner portraits?

alongcaca

Quote from: DatKofGuy on April 08, 2026, 12:14:50 AM
Would this method work for trying to replicate character portraits? Like say KOF2002 or 2003 winner portraits?
It's certainly possible; you can test it yourself. However, due to the differences in size, you'll need additional skills in image editing software. You shouldn't expect to get the desired result with just one prompt.