7:5 and 5:7 images: crop from 3:2, 4:3, 3:4 or 2:3 on Sume

If you need 7:5 or 5:7, generate at the nearest listed ratio and crop. FLUX 3 Image is not listed in the Sume image catalog (read 2026-10-09). No Sume row lists 7:5 or 5:7, so they are not among the documented values. From 3:2 you lose about 6 percent of the width, and from 4:3 about 4 percent of the height.

Which ratios does BFL list near 7:5?

BFL puts 7:5 in its landscape photo group with 3:2 and 4:3, and 5:7 in the portrait group with 3:4 and 2:3 (BFL docs, read 2026-10-09). Those four neighbors are all on Sume.

How many pixels does the crop cost?

Using Nano Banana 2.1 2K frames from Google's size table (read 2026-10-09), the crop keeps the full edge on one side and trims the other:

Which neighbor should I start from?

Start from 4:3 or 3:4: the trim is 78 px at 2K, about 3 percent of a long edge, which loses less of the composition. The 3:2 and 2:3 crops cost about 6 percent. Say the subject should sit in the middle in the prompt, since the crop removes the edges.

The scale is done in your code. Sume does not apply target_pixels, and it returns the pixels the row produces.

• Compose with a safe margin of at least 5 percent.
• Keep text and logos away from the cropped edges.
• Verify the result size after the crop, not before.

Can the crop be done in code?

Yes: the arithmetic is two lines. This helper prints the output size for a width, height and target ratio.

If 7:5 is a hard requirement for a print order, the cleanest route is to generate at 4:3 or 3:2, crop once and keep the full-size original. A second crop later, for another channel, should start from the original, not from the cropped file. Name the files with the ratio so a 7:5 never gets mistaken for a 3:2.

def crop(w, h, rw, rh):
    if w * rh > h * rw:      # frame too wide
        return round(h * rw / rh), h
    return w, round(w * rh / rw)  # frame too tall

print(crop(2528, 1696, 7, 5))
print(crop(2400, 1792, 7, 5))
print(crop(1792, 2400, 5, 7))
print(crop(1696, 2528, 5, 7))