Using ImgtoImg to create a new Image
By Cayla Catz
Using Image to Image to create a new image What I will show you here Getting your First Image Img2Img with 1 starter image and a text prompt Img2Img with 2 starter images and a text prompt Img2Img with 3 starter images and a text prompt I will show you the results in 5 models: P-Image, GPT 1.5 Low, Grok, Nano Banana and SeeDream 4.5. The challenge What kickstarted my choosing to use a start image in image to image creation rather than text to image creation was Cheinia @Cheinia had issued a QT challenge to fill a white box in this image. https://budgetpixel.com/p/14105 Getting your First Image I couldn't figure out how to see his prompt although I'm pretty certain it's an open prompt as I can see part of the prompt when people link to Cheinia's post. Regardless, there are 2 easy ways to get Cheinia's initial starting image. 1. Easiest. Right click and download Cheineia's image. Do not download someone's image unless they give permission. However in a challenge, when someone says "do this with this picture", I'm going to assume that's permission. 2. Or Recreate the prompt. I used GPT Low at 3:2 with this prompt I made. Feel free to use this prompt to make a similar picture. If you're using a different model, you may have to fiddle with it some. FYI: Squircle = a shape halfway between a circle and a square. It's a real word. an astronaut stands on a desolate moonscape on a starfilled night, gazing at a massive glowing squircle resting on the ground filled with white light filling more than 1/3 of the image with the big giant white box Img2Img : 1 image + prompt text One way to use Img2Img is to use one image and then use the prompt to say what you want changed. You can use more than one picture but for now we are using only the one. Bring in a picture by 1. upload from your computer/phone 2. or use an image you've made on BP under assets As you can see I've already loaded the image I made from my prompt above. Now in the prompt box, you say what you want to change in the picture. I made a very basic prompt: fill the white space with the Jurassic era and dinosaurs Since the box is being filled by an image generated by the text prompt, the image inside the box will be different but on the same subject. P-Image result GPT 1.5 Low result: Grok result: Nano Banana result: Seedream 4.5 result: You can see the different models mostly kept my image as it was although a few moved the camera closer. Seedream was the only one that added a border to the white square. P-image had some dinosaurs spillout onto the moonscape. I didn't tell the prompt to do that. You can tell a prompt to spill out or add a border if you like it. Img2Img: 2 images (or more) + prompt text Another thing you can do is put two or more images together. I added a second picture from my assets I added a simple prompt: Put Image 2 into the white box on the moon in Image 1 Because they should be using my second image, the results should look pretty similar. That's pretty much what happened except for P-image and SeeDream 4.5. Warning: If you are switching between models like I am, keep an eye on the size. Some will match the size of the first picture but some models will switch you to 1:1. I wanted to stay at 3:2 so i kept having to switch it back to 3:2. P-image result GPT 1.5 Low result Grok result Nano Banana I got this Seedream 4.5 result: Img2Img: 3 images (or more) + prompt text Different models allow different amounts of total images you can use in Img2Img P-image allows 3 pics Grok allows 3 pics Gpt 1.5 allows 6 pics Nano Banana (all models) allows 6 pics SeeDream 4, 4.5 and 5 allows 6 pics Flux Dev, Kontext Max, Kontext Pro, Krea Dev allows 2 pics Flux 1.1 allows 2 pics Flux 2 Max allows 2 pics Flux 2 Klein allows 3 pics Flux 2 (Dev, Flex, Max, Pro allows 4 pics Ideogram (all models) allows 3 pics Minimax allows 1 pic Qwen image 2, 2 pro, image edit, Image edit plus allows 3 pics TenCent Gen 4 allows 3 pics TenCent Hunyan Image 3 Instruct allows 2 images Wan 2.5 & 2.6 allows 2 images So next we are going to try 3 images: the astronaut image and 2 others with P-image, Grok, Gpt 1.5 Low, Nano Banana, SeeDream 4.5. As before you just add them into the appropriate places Image 2 Image 3 This is trickier as there is more room for the AI to misinterpret what you mean. I find 2 images are fairly easy for me to get the result I want in merging elements from the two images. Using 3 images, I tend to burn credits getting the prompt where I need it. You ae going to see very divergent responses this time. This is my prompt: Put image 2 inside the white box inside image 1. Put the flying cat in Image 3 in the box to on the left side flying towards the black cat in Image 2. P-image Result (failed but it's an interesting fail): GPT 1.5 Low. I felt GPT was pretty successful at getting what I described in my prompt. Grok result: Nano Banana. this one is pretty good too. Seedream 4.5. Again Seedream decided to play with my instructions a bit Conclusion It can be a lot of fun to merge images. For example maybe I want the flying cat to have tea with this cat. Put the flying cat first as it's 16:9 so it's the larger picture. Then I added the 2nd picture which was 1:1. Then I added my prompt " Put the flying cat in Image 1 into image 2 by sitting down to have tea together with the dragoncat. " And SeeDream 4.5 gave me this and there they were, having tea. I tried the prompt again with GPT and asked it to keep the 2nd images art style. "Put the flying cat in Image 1 into image 2 by sitting down to have tea together with the dragoncat. Keep the art style of Image 2." It came out pretty good. So as you can see there is a lot of fun you can have merging pictures together.