ElanoryaJewelry AI
Why your result comes back in a different shape
The frame of a generated image is chosen from a fixed set of five ratios. Your photo's ratio is measured and rounded to the nearest one of them, so a base photo that sits between two ratios comes back as the nearer one. Nothing in the flow asks you which frame you wanted.
What decides the shape of the result
The frame of a generated image is not carried over from the model photo you supplied. Before the run is sent, a width-to-height ratio is measured and compared against five fixed values — 1:1, 3:4, 4:3, 9:16 and 16:9. Whichever of the five is nearest is the ratio that gets requested for the output.
That is the whole mechanism. A base photo standing exactly on one of the five is requested at that ratio. A base photo standing between two of them is requested at the nearer one, and the distance it moves is however far it was from that ratio to begin with. There is no sixth option and no way to ask for a ratio outside the set.
The rounding applies to the number that is asked for, not to your photograph. Your model photo is not cropped to fit one of the five and it is not stretched to fit one. What changes is the shape of the frame that comes back.
The five frames a result can come back in
These five are the complete set. Any submission is requested at whichever one of them is nearest to its measured ratio.
| Requested ratio | Shape of the frame | Longer side |
|---|---|---|
| 1:1 | Square | Neither |
| 3:4 | Portrait | Height |
| 4:3 | Landscape | Width |
| 9:16 | Tall portrait | Height |
| 16:9 | Wide landscape | Width |
What the rule touches, and what it leaves alone
The rounding decides one thing. It is worth being exact about what that thing is, because several nearby questions look like the same question and are not.
- It decides the ratio requested for the output frame.
- One of 1:1, 3:4, 4:3, 9:16 or 16:9 is chosen and that is what the run asks for.
- It does not crop or stretch the photo you submitted.
- The measurement is read from your photo; the rounding is applied to the request. Your file is not reshaped by this rule.
- It does not decide how large the result is.
- Output resolution is fixed rather than chosen, and it is settled by a separate rule with its own page. Ratio and size are two different decisions.
- It is not something you can set.
- The aspect-ratio option list is dead code — it is present in the source but never rendered. There is no picker, no default to change and no match-original switch.
- Orientation is corrected before anything else.
- EXIF orientation is applied to the pixels and the orientation metadata is removed, so a photo that was recorded rotated is straightened rather than left inconsistent.
Deciding the frame before you submit
Everything about the output frame is settled by the photo that goes in. That places the entire decision before the run, in this order.
Name the frame you need
Decide which of the five the finished image has to be: 1:1, 3:4, 4:3, 9:16 or 16:9. Deciding after the result arrives is too late, because there is no control that changes the frame of a result you already have.
Bring the base photo to that ratio
Prepare the model photo at the ratio you named before you bring it into the app. The measured ratio of the file you submit is the only input the rule reads.
Check the body region survived
Confirm that the region the piece belongs on is still fully inside the frame — the hand for a ring, the wrist for a bracelet, the neck and upper chest for a necklace, the ear for an earring. A crop that improves the ratio and loses the region has cost you the run.
Submit once
The ratio is measured and rounded as the request is prepared. A submission is queued for a single attempt, so one submission is one provider run — there is no second pass that could come back framed differently.
Compare the returned frame with the one you planned
If the shape is not the one you wanted, the correction belongs to the crop rather than to a setting. Adjust the base photo toward the ratio you need and submit that instead.
What can go wrong when you crop toward a ratio
Cropping is the only lever you have over the frame, and it is a lever that can damage the photograph it is applied to.
The crop reaches the target ratio but cuts into the body region the piece is placed on.
Check the region for the category before you accept the crop: hand for a ring, wrist for a bracelet, neck and upper chest for a necklace, ear for an earring.
A wide photo is cropped hard toward 9:16 and the hand or wrist is pushed to the frame edge.
For a tall finished frame, start from a photo that is already tall enough to hold the region comfortably rather than cutting a wide one down to it.
Pieces in one catalogue are cropped individually and the set comes back in several different frames.
Choose a single target ratio for the set first, then crop every base photo toward it. Each submission is measured on its own, so consistency has to come from the crops.
A base photo sits almost exactly between two of the five, and small differences in preparation send near-identical photos to different frames.
Crop deliberately toward one of the two rather than leaving the photo near the midpoint, so the outcome does not depend on a marginal difference.
What this rule does not do
- It decides a ratio and nothing else. It does not decide how large the result is, and it does not decide anything about the file that is returned.
- The five ratios are the complete set. Any measured ratio is requested as the nearest of the five, so a result cannot come back at a ratio outside them.
- There is no aspect-ratio control: the option list exists in the source but is never rendered, so there is nothing to configure and no match-original choice.
- The 3:4 assumption used when measurements are unusable is a fallback inside the rule, not a default you set and not a preference the app remembers.
- A submission is queued for a single attempt, so a run cannot quietly repeat itself and return a differently framed image.
- Output resolution is fixed rather than chosen; that is a separate rule and it has its own page.
What this page does not cover
How large a generated image is, and why its size is fixed upstream instead of offered as a setting, is covered separately. This page is only about the shape of the frame.
When credit is deducted for a run, and what the confirmation states before a run starts, is also covered separately. Nothing in the framing rule changes when a charge happens.
Which image files the app accepts, and why a photo can be refused before any of this applies, is a different question with a different answer. So is scale — how large a piece appears within the frame — which varies by category and is not decided by the ratio rule described here.
Questions
Why is my generated image a different aspect ratio than the model photo I gave it?
Because the frame is not inherited from that photo. Its width-to-height ratio is measured and rounded to the nearest of five fixed ratios — 1:1, 3:4, 4:3, 9:16 and 16:9 — and the rounded ratio is what is requested for the output. Unless your photo was already at one of the five, the result comes back at the nearest one instead.
Is there a setting that keeps the original shape?
No. An aspect-ratio option list exists in the source but is never rendered, so there is no picker, no default to change and no match-the-original option. The ratio of the base photo you submit is the only input to the decision.
Does the app crop or stretch my model photo to fit one of the five ratios?
No. The rounding is applied to the ratio that is requested for the output, not to your photograph. Separately, EXIF orientation is applied to the pixels and the orientation metadata is removed, so a photo recorded rotated is straightened rather than left inconsistent.
What happens if the dimensions of my photo cannot be read?
The run is not held back and the ratio is not left undecided: 3/4 is assumed and the request goes out at 3:4. If a result comes back at 3:4 when the base photo was clearly another shape, an unreadable measurement is worth ruling out first.