Zoom in on a small product in frame: AI recompose or crop and upscale

When a product is small in the frame, you have two routes on Sume. The first is to crop the area in Pillow and, if the crop is too small, enlarge it with POST /v1/image-upscale-1.0/upscale, which is $0.20 per image and keeps the real pixels as far as an upscaler can. The second is to ask ideogram/ideogram-v4.5 to recompose the photo with the product larger, which costs $0.075 at medium but redraws the product, so label text and small details can change.

Ideogram's 4.5 page lists zoom editing among its edit examples. That makes the recompose route a real option for lifestyle shots, but not for photos where the product has to be a faithful record.

Route 1: crop, then upscale

A crop is lossless in the sense that every pixel left is a pixel the camera captured. If the product occupies the middle third of a 4000 pixel wide photo, a crop to that third is still over 1300 pixels wide, enough for many web uses without any model. When you need more, the Sume upscale route takes a public HTTPS image_url, an upscale_factor from 1 to 4 (the default is 2) and an output_format of png, jpg or webp.

The catalogue price is $0.20 per image, which Sume prices as a reserved estimate of about 16 output megapixels. It is the most expensive step in this comparison, so crop first and enlarge only what you will actually publish.

from PIL import Image

im = Image.open("shelf.jpg")
w, h = im.size
box = (int(w * 0.30), int(h * 0.25), int(w * 0.70), int(h * 0.80))
crop = im.crop(box)
crop.save("product_crop.png")
print("crop size", crop.size)

Route 2: recompose with an edit

Send the photo as the first input_references entry and say what you want: the same product, filling about 60 percent of the frame, same background style, same lighting. Without aspect_ratio the edit keeps the source shape, so the new framing happens inside the original canvas. The model draws a new picture around what it believes the product looks like.

This is where drift matters. A label that read a clear brand name in the original can come back with a similar but wrong word. For a catalogue, record-keeping or marketplace image where the product must be accurate, use route 1.

Cost and decision

Both routes are priced per image on Sume. The crop is free; the upscale and the edit are billed.

Resolution and framing details

Before you pick a route, measure what you have. A phone photo is often 4000 pixels wide, so a central crop is still large; a marketplace image saved at 1000 pixels is not. Ideogram says 4.5 handles edits up to 4,016 x 6,016 pixels without downsampling on its own page, but on Sume the edit keeps the geometry of the source when you omit aspect_ratio, and resolution on its own does not change a source size. That means a recompose edit will not give you more real pixels than the source has.

The practical order is: decide the final display size, crop to the framing you want, check whether the crop meets that size, and only enlarge when it does not. If the framing is wrong because the subject is off-centre rather than small, a crop alone solves it. If you must keep the original canvas, ratio and all, a recompose edit is the only way to move the product within it.

A short decision rule

If the product label has to be read, crop. If the crop is under about 1000 pixels wide and you need more, upscale. If the picture is a mood image, a social post or a hero where the look matters more than a faithful label, recompose. If you are unsure, do both on three photos and compare labels at full size.