The Cheek Mark Stayed. The Close-Up Told Me More Than That.

A wider generated portrait and a generated facial close-up, compared against a plain crop: which details survive, which are new, and how to keep those two observations apart.

The first thing I looked for in the close-up was the little dark mark on the right side of the picture. It was still there. In the wider portrait it sits below the cheekbone, and in the close-up I could find it in the same part of the face. At first that felt like enough to call the edit consistent. Then I cut a square directly out of the wider portrait and put it beside the generated close-up. The two views made it much easier to see which detail had survived and where the edit was giving me new information I couldn’t verify from the source.

The wider generated portrait beside the generated facial close-up

A face with one place to look

I made a fictional portrait of a bookbinder in a blue jacket. The wider frame shows her head and shoulders, with shelves behind her and soft light from the left. For this comparison I wanted one small feature I could reliably find again. I specified a single mark on her left cheek, which is on the viewer’s right. The generated reference included it clearly enough that I could point to it in the original.

The close-up had a different job. I wanted both cheeks in the frame, with the viewer close enough to read the expression and the surface of the skin. The larger face also made a small change easier to see. A feature that barely registers in a profile card can become the first thing someone notices in a close portrait.

I used the image editing mode in Grok Image 2.0 and uploaded the 1024 × 1024 reference. The editor was set to Standard and showed 2K based on the source image. I made one edit and downloaded the 2048 × 2048 result. Here is the instruction I submitted:

Reframe this exact fictional woman into a tight editorial facial close-up, from just above the eyebrows to just below her chin, with both cheeks visible. Preserve the same person, neutral expression, left-side diffuse daylight, natural skin texture, and the existing small dark mole below the cheekbone on her left cheek (viewer’s right). Keep that mole on the same side and in the same position relative to her eye and nose. Do not add freckles or additional moles. Square framing.

Completed editing workspace with the source, prompt, settings and resulting close-up

The crop gave me a second way to look

I made a 419 × 419 crop around the face in the original. The crop came straight from the reference file. For the comparison image I displayed the crop and the generated close-up side by side in equally sized panels, with no extra sharpening. The framing and apparent face size aren’t an exact match. I used the panels to compare visible features by eye.

A conventional crop of the original beside the generated close-up

The clearest point of agreement was the cheek mark. I could trace it below and outward from the eye in both images. The light still came across the face from the left, and the expression stayed calm. The generated face was sharper at this display size, particularly around the skin and eyes. Fine lines and pores were easier to see, and a few faint points below the main mark drew my attention in the larger view. The reference crop didn’t give me enough detail to say whether those faint points corresponded to anything already there.

I would put the large cheek mark and the finer points in different review notes. I can say I found the main mark again. I can also say the close-up presents finer texture. The wider starting image gives me no reliable way to check each tiny point. That’s where the distinction matters if someone wants the close-up to stand for the same portrait.

The crop also helped me separate a framing decision from an editing decision. The reference already gave me room to trim away the jacket and most of the shelves. If that’s the only change a layout needs, the crop is available immediately. The tighter generated image becomes useful when the face itself has to read at a larger size. At that point I’d pay particular attention to the cheek beside the nose and the area below the eyes. Those regions are easy to compare quickly, and they carry enough fine detail that a sharper rendering can start to look more certain than the reference warrants.

Which one I’d use

For a fictional editorial portrait, I’d consider the generated close-up. It gives me a tighter composition and a readable face, and it keeps the feature that helps connect the two views. I’d check the final page at its actual display size before choosing the image. In a small layout, the difference between the original crop and the generated close-up might matter less than how the face sits next to the text.

For a picture whose small details need to match an approved reference, I’d keep the conventional crop or ask for a closer source image. My first review pass would identify two features I can locate in both images and compare their positions relative to the eyes and nose. My second pass would flag anything visible in the new picture that the original is too soft or too small to confirm. Keeping those observations separate makes it possible to say what the edit did well and where I would still want a better reference. In this run, the cheek mark was easy to verify. The finest skin texture was the part I would leave open.


Write a comment