How To Get Good Quality Complex Images - Part 2

In the first part of this tutorial we have seen how to get a complex image from text, noting how, with the same prompt, different models can create very different images. So users need to learn which models are best suited to create the complex images they have in mind, otherwise they will waste credits and time. The same can be said for the process of improving or enhancing images through image-to-image models: the results can be unsatisfactory and sometimes it is not easy to find the right model to achieve what we want. Now I will show you some examples of image enhancement, first displaying the original image, called Input Image or Reference Image depending on the model used, then the enhanced image, with a short comment on the model's performance. The prompt I used was essentially the following: " Improve and enhance the reference image making it less cartoon-like and more photorealistic, add more details, matching those already present. It must look like a paradise dreamworld with phantasmagoric atmosphere, amazing digital art with lots of details, sharp focus, warm cinematic light, HD, masterpiece, award winning photo. " Variations of the prompt could include " Merge the reference images... " with those models that allow to use more than one reference image (with caution, because when there are more than two images the model often focuses only on the first one), or " increase contrast... " for those images which are somewhat faded. So, let's start. Here Nano-Banana 2.5 didn't produce any improvement: just a waste of credits 👎! I tried also Nano-Banana 3: same result 👎. With the input image above, Flux Kontext Dev (25 credits) produced some improvements in image focus and sharpness. Flux Kontext Max (80 credits), on the other hand, produced a more vintage and warm photorealistic result. Not exactly what I wanted, but interesting 👍🏼 P-Image Edit (15 credits) is a cheap model that can give amazing results, as you can see in these colorful fancy images 👍🏼👍🏼! Here is an example of how the same image was modified by Flux 2 Flex (40-80 credits), which has some adjustable parameters which the user can set 👍🏼. SeedEdit 3 (30 credits) improved a bit the image. This is the effect of SeeDream 4 (33 credits), which always tends to produce a darker, more contrasty image. Instead this image, created with SeeDream 4, has not been significantly improved by SeedEdit 3 👎. P-Image Edit (15 credits) tried to add some details to this image created with Z-Image Turbo. Some good results, with the same image, were obtained with Flux 2 Max (80 credits, first image) and Flux 2 Flex (80 credits, 2nd and 3rd image) 👍🏼👍🏼. Two more interesting transformations obtained with Flux 2 Flex. In the end, I'd like to point out that the same models can give different results on different platforms: all these images were obtained using BudgetPixel's models.