The red coat came through. The pants went pale.
We started with a studio portrait of a man standing beside a covered table. The second photo showed a red military coat from The Met. The result put him in a red coat with a black front and rows of gold fastenings. The hand beside the table and the sweep of the curtain were still easy to recognize.
Left to right, the portrait we started with, the front view we supplied as a clothing reference, and the saved Nano Banana 2.1 result.
It would be easy to stop at the red coat. It's the biggest change in the image, and from a distance it looks at home in the old photograph. But the usefulness of the result depends on what you wanted to change. If you're trying out an outfit, the whole figure is in play. If you're deciding whether one coat works with the dark pants you already picked, those pants are part of the comparison.
The reference looked straightforward in the small preview. It got more interesting when we opened the museum's other photographs of the coat.
What the two pictures actually show
The portrait supplies the person, pose, and setting. Look at the pants and the hand beside the table before opening the edited version. Library of Congress portrait
For a full-length clothing edit, I'd pick a portrait that shows the shoulders and arms as well as what sits below the garment. A crop at the chest gives you a clear face but leaves you guessing about the way the jacket meets the pants. Here, the source has room to see the original jacket hem, both shoulders, and the full length of the pants. The table and curtain are good landmarks if you need to compare the framing later.
This front view was the clothing reference. The pale area below the red front is easy to read as part of a whole outfit at thumbnail size. The Met's military coat
The front photo has most of what you'd want for this edit. The collar, the black panel down the chest, and the gold fastenings are plainly visible. Lower down, the pale fabric splits into two shapes. The first version of my notes called those shapes trousers. The Met catalogs the object as a military coat, and the rear photograph gives you a better look at its split tails and the pale fabric along them.
The rear view is useful for checking the garment's shape. This photo was consulted afterward and was never sent to the model. The Met's rear view
The saved result shows pale pants. What led the model there is open. It might have taken the shape below the waist as a cue or supplied pants that went with the coat. For your own files, open any alternate view you have before deciding what to crop. A crop at the red waistline would cut off details of this particular coat. A clearer front photo that shows the whole item would give you a better starting point for a coat-only edit.
Google's virtual try-on interface gives the person and garment separate input fields. If your editor uses numbered slots, check the order before you write “the first photo” or “the second photo.” A reference can be the second image here and the first image in a different workflow. Google Cloud's virtual try-on example
The setup behind the saved result
To follow the same two-upload workflow, open Image tools from the homepage and choose AI Clothes Changer. Image 2 Image
Choose Use a garment photo and put the portrait in Person photo. Put the coat's front view in Garment photo. This workflow requires signing in and choosing a model that accepts two reference images. We used Nano Banana 2.1 for the saved run. The result was a portrait-shaped 816 × 1280 JPEG. You can download both sources directly if you want to repeat the setup. Portrait file · Garment file
In that mode, the editor sends the portrait first, the garment second, and this instruction with them.
Change the outfit of the person in the first image to the garment shown in the second image. Use the second image only as the clothing reference. Match its color, cut, pattern, and fabric. Fit it to the person's pose and the original lighting.
We also sent a separate instruction about what to keep.
Keep the face, hair, body shape, pose, and background.
The outfit is the thing it asked to change. The keep list covers the person and the scene. A coat-only brief gives the pants their own instruction. If you're in this preset and want one garment, the second upload is the place to start. For control over the wording, use an editor with a prompt box and support for two references.
How I'd run a coat-only test
In an editable image-to-image workflow, put the portrait first. Use the same front view as the second image initially, even if you suspect it caused trouble. That way, you can see what changes when the instruction gets more specific. I'd try this wording.
In the first photo, replace the jacket with the red military coat from the second photo. Keep the dark pants, the standing pose, the table, and the camera framing from the first photo. Keep the coat's black front and gold trim.
I'd start all three runs in the same editor and use the same model and output settings. If it lets you set a seed, record it. Keep the original portrait as the first input each time. Reusing the last generated image can carry its changes into the next attempt, which makes it harder to tell what the new prompt did.
| Run | Clothing photo | Instruction | What the comparison can tell you |
|---|---|---|---|
| A | Original front view | Outfit instruction and preserve line from the saved test | Establish a starting point in your editable editor. |
| B | Same front view | Coat-only wording above | See what changes when the wording narrows the job and names the pants. |
| C | A clearer, single-garment photo | Same text as B | Look for the effect of changing the second image. |
These are proposed runs. The triptych shows the earlier outfit edit. If you only have time for one follow-up, try B and check the waist. Run A and B in the same editor if you want a fair comparison of the wording. If the pants still come out pale and you can get another clear front photo of the same item, try C. I'd stick with the same coat for all three runs so its shape and details give you the same things to check each time.
If the specific line about the pants is what you want to test, make two copies of the coat-only prompt. In one, remove only the words about the dark pants. Leave the coat, pose, and framing instructions alone. That smaller comparison is useful if you already have a decent coat swap and want to know what moved the color below the hem.
Look at the waist before you look at the buttons
Start with the whole image on screen. Check the waist and the legs against the original, then look at the hand near the table. If the sleeve swallows the hand or the elbow moves away from the table, the coat may sit awkwardly on the person. Use the curtain and the table edge to spot a change in framing.
Put both pictures at the same display height as you do this. The prepared portrait in our test was 510 × 800, and the saved result was 816 × 1280. They have the same proportions. At the same display height you can compare the position of the shoes and the table edge without the larger file looking like a tighter shot. If your own export crops the shoes or pushes the coat against the top of the frame, check the aspect-ratio setting before you spend another run changing the prompt.
Then look at what came from the clothing photo. The museum front is a bright scarlet. In the edited portrait, the coat reads as a muted, brownish red against the sepia setting. I like it for this portrait. For a catalog shot, I'd judge the color against the garment under the lighting the catalog uses. The same shift can work for one image and make the other one misleading.
The cuff detail makes the shape of the leaves and the narrow bands of braid easy to see. It was used for visual review only. The Met's cuff detail
I saved the cuff for last because the large changes are easier to settle first. Gold on a black cuff is enough to make the coat look familiar at a glance. The leaf shapes and bands of braid give you something more specific to check if the exact garment matters. If you run the edit yourself, open your full-size result beside the museum detail and see how much of that pattern survived.
For the next coat-only run, I'd keep one small note next to the file. “Dark pants still visible below the coat. Same hand beside the table. Black front and gold fastenings readable.” It takes a few seconds to write and gives you something more useful than a folder full of images called coat-final.
I worked with AI on this article. I set the direction and gave editorial feedback, and AI drafted and revised the English text. The opening comparison comes from a recorded project test. The coat-only prompt and three-run comparison are proposals for a follow-up.





