Recolouring An Image, The AI Way

Introduction I made an image in Flux 2 Pro the other day. I thought that it seemed like a good image, except for the washed-out, hazy look I've been seeing so often from that model. I didn't notice the mistakes on the dragon until much later because I was focused on the wizard. The Reference I remember @Sealine dealt with a similar problem to great effect, so naturally, I asked for help. Notice that the original image on the left is a beautiful work, but it exhibits the same washed-out look. Her approach was remarkably simple and clearly effective. She used Nano Banana image-2-image to fix it with this prompt: Keep image the same, but change to a brighter color palette with no sepia tone. I had been convinced that there was some complex external process involved in her getting such a fantastic result. I encourage you all to check out @Sealine 's other work (all of it fantastic) at her BudgetPixel page . Steps To Fix My Image Being an engineer, I like to approach everything in small, incremental steps. Changing as little as possible at each step to control the results and ensure that I'm more likely to understand them. 1. Do what @Sealine did. I copied her prompt word for word: Keep image the same, but change to a brighter color palette with no sepia tone. And referenced my original image. Nano Banana did precisely what the prompt told it to. The colour palette was definitely brighter, and as a result, the image already looks a lot better (minus the dragon mess, lol). Unfortunately, I still didn't like it. It was still washed-out and hazy, just brighter. 2. Add contrast to improve the washed-out look. Having a non-artist brain, I thought that perhaps what I needed was to somehow introduce some contrast into the mix. I modified the prompt to: Keep image the same, but change to a brighter and more contrasting color palette with no sepia tone. It definitely did that (perhaps overdid): 3. Pull back the contrast I was startled by how much contrast it added. I immediately jumped to the conclusion that it was too much. Now that I look at it again, I think this could also become a good image with a little more work. Perhaps I will experiment a little more to see what comes out of it. The next attempt was targeted at reducing the amount of contrast I wanted to add. I changed the prompt and tried again: Keep image the same, but change to a brighter and slightly more contrasting color palette with no sepia tone. 4. Fixing the image sharpness I concluded that my problem was no longer about being washed out. Perhaps it never was. I decided the previous image wasn't sharp enough, and fixing it turned out to be relatively straightforward. Keep image the same, but change to a brighter and slightly more contrasting color palette with no sepia tone, and improve the sharpness . In my opinion, this resulted in a good image. (aside from the dragon issues, which I will fix separately). Conclusion I think the takeaway here is that there is no single, fix-all approach. No magic pills. What matters most here is that I was able to work with a fellow artist, starting with a technique she worked out, and using it as a jumping-off point for the fix I needed. I'm happy with the results. Thank you, @Sealine , for the collaboration. I appreciate it. Errata and Reality In a perfect world, it would have gone exactly as I described it there. The approach is sound and controlled. Make simple, small changes one at a time, which lets you: • control the output, • understand each change, • match the output with your expectations, • make better decisions on how to proceed at each step. The reality is, I'm human, and due to several reasons, including impatience, I got frustrated because I wasn't seeing the results I wanted. None of the images was blowing my socks off. I deviated right after step 3, where I pulled back the contrast. I rewrote the prompt from scratch, leading with the changes I wanted and then outlining my constraints. It's the way I usually perform image edits. The reason I did it here was that, in my head, I thought going with that approach might improve the image, since the change part of the prompt was higher up in the model's focus (see the article about prompting technical details, which I have not yet written). At the same time, the idea for sharpness came to me, so I added that snippet to the prompt as well. Use a brighter color palette with no sepia, and slightly enhance the contrast and the sharpness. Keep all other aspects of the original image exactly the same. Make no other changes. Yeah, the sharpness helped a lot. The image was definitely better at this point, but I didn't like it. Increasing the sharpness seemed to mute the colours a little. Frustrated and annoyed, I decided to change the model. I thought that Nano Banana Pro might do a much better job than Nano Banana. The one thing I goofed on was that changing to the new model forced me into a 1:1 ratio because Nano Banana Pro does not seem to support a 'Match image ratio' setting. I do quite like this image (minus the dragon issues). I think that Nano Banana Pro lent some weight to cleaning up the entire image. Yes, I think it did turn out better than the one I did in step 4. I ran one more edit, this time forcing Nano Banana Pro to 9:16, which is the same as the original image. Final Images