Yes — AI can put people from separate photos into one group photo, and the best models in our test kept every person and their face in all six scenes. We gave eight models the same photos of three to six people, one photo each, plus one existing group photo to add a person to, and measured every face against its input. Where the models differ is how faithfully they keep faces, and whether they keep everyone.
- Best overall and best value: Seedream 5.0 Flash — all six usable, five passed every check, 15 credits a photo.
- Best faces: GPT Image 2 — all six usable, five passed every check, the highest face match, at 50–75 credits for a group.
- Watch out: Nano Banana Pro left a person out of two photos, and Seedream 5.0 Pro swapped someone out when asked to add a person.
The results
| # | Model | Usable | Passed every check | Face match | Credits |
|---|---|---|---|---|---|
| 1 | GPT Image 2 | 6 of 6 | 5 of 6 | 0.88 | 61 |
| 2 | Seedream 5.0 Flash | 6 of 6 | 5 of 6 | 0.86 | 15 |
| 3 | GPT Image 2.5 | 6 of 6 | 4 of 6 | 0.87 | 37 |
| 4 | Nano Banana 2 | 6 of 6 | 2 of 6 | 0.79 | 40 |
| 5 | FLUX 3 | 6 of 6 | 2 of 6 | 0.78 | 25 |
| 6 | Nano Banana 2.1 | 5 of 6 | 4 of 6 | 0.79 | 30 |
| 7 | Seedream 5.0 Pro | 5 of 6 | 2 of 6 | 0.76 | 35 |
| 8 | Nano Banana Pro | 4 of 6 | 0 of 6 | 0.69 | 75 |
Usable counts the photos that failed no check and Passed every check those that met all four in full; nothing is averaged. A photo with anyone left out or anyone added fails the first check, so it is never counted as usable. Face match compares each face in the result with its input photo, matched person by person (1.00 is identical). Credits are the average over the six requests; the price of some models rises with the number of photos you add.
How we tested
Every model got the same photos and the same description, once each. The people are AI-generated, so no real person's face was used.
- Tested
- 2026-10-15, on PicYep, as a visitor would use it
- Models
- Nano Banana 2.1, Nano Banana 2, Nano Banana Pro, Seedream 5.0 Pro, FLUX 3, GPT Image 2.5, GPT Image 2, Seedream 5.0 Flash
- Settings
- 1K, 4:5, one photo per request, each model's default quality, the description sent as written with no rewrite
- Size of the test
- 6 cases, 48 requests, one each, every result kept
What each photo was checked for
- Everyone is there. Every person or pet in the description appears once: nobody missing, doubled or added.
- Looks like the input. Each person (or pet) is recognisably the one in their input photo: face, hair, build.
- As described. Scene, clothes, pose and style follow the description.
- Realistic. Passes as a photo: skin, hands, eyes and light have no obvious AI faults.
Each check is marked passed, partly or failed; no points are added up or averaged. A photo is usable when it fails no check, and the table counts the usable photos and the photos that passed every check. A request that ended without a photo counts as not usable. Every result is shown below, each with a link to its original file.
Face match is measured, not judged: FaceNet (InceptionResnetV1 trained on VGGFace2) turns each face into numbers, and we give the cosine similarity between the face in the result and the face in its input photo, person by person. 1.00 is the same face; our different input people score -0.22 to 0.22 against each other.
The descriptions name people by the order of their photos, the way we recommend on the group photo page — for example: "The woman from the first photo, the man from the second photo and the woman from the third photo, friends having dinner together…".
Photo by photo
Three friends at dinner
Every model put all three at the table with string lights and raised glasses. The GPT models kept their faces closest; the others drew looser likenesses.
The woman from the first photo, the man from the second photo and the woman from the third photo, friends having dinner together at a long wooden table on a patio, string lights overhead, laughing and raising their glasses, warm evening light, a candid photo, realistic skin texture.
Nano Banana 2.1Original
Usable · partly: likenesslikeness weak
Nano Banana 2Original
Usable · partly: likenesslikeness weak
Nano Banana ProOriginal
Usable · partly: likenesslikeness weak
Seedream 5.0 ProOriginal
Usable · partly: likenesslikeness weak
FLUX 3Original
Usable · partly: likenesslikeness weak
GPT Image 2.5Original
Passed every check
GPT Image 2Original
Passed every check
Seedream 5.0 FlashOriginal
Usable · partly: likenesslikeness weak
Four people at a holiday party
All eight kept the four people together in front of a tree. FLUX 3 sat them on a sofa instead of standing.
The four people from the four photos at a holiday party in a cozy living room, standing together in front of a decorated Christmas tree, holding mugs of cocoa, smiling at the camera, warm lights, realistic photo.
Nano Banana 2.1Original
Passed every check
Nano Banana 2Original
Passed every check
Nano Banana ProOriginal
Usable · partly: likenesslikeness weak
Seedream 5.0 ProOriginal
Usable · partly: likenesslikeness weak
FLUX 3Original
Usable · partly: likeness, as describedscene not as described · likeness weak
GPT Image 2.5Original
Passed every check
GPT Image 2Original
Passed every check
Seedream 5.0 FlashOriginal
Passed every check
A team of five
All eight results have all five people. GPT Image 2.5 cropped them into a close huddle rather than a standing team photo.
A team photo of the five people from the five photos, standing together in a bright modern office in smart-casual clothes, smiling at the camera, soft natural light, realistic photo.
Nano Banana 2.1Original
Passed every check
Nano Banana 2Original
Usable · partly: likenesslikeness weak
Nano Banana ProOriginal
Usable · partly: likenesslikeness weak
Seedream 5.0 ProOriginal
Passed every check
FLUX 3Original
Passed every check
GPT Image 2.5Original
Usable · partly: as describedscene not as described
GPT Image 2Original
Passed every check
Seedream 5.0 FlashOriginal
Passed every check
Six friends on a picnic
The hardest case. Nano Banana Pro left one person out and turned another away from the camera; the rest kept all six. In wide candid shots like this, faces are small, and face match drops for every model.
The six people from the six photos, friends on a picnic in a sunny park, sitting together on a blanket with a basket of food, laughing, summer afternoon light, a candid photo, realistic skin texture.
Nano Banana 2.1Original
Not usable · failed: likenesslikeness weak
Nano Banana 2Original
Usable · partly: likenesslikeness weak
Nano Banana ProOriginal
Not usable · failed: everyone there, likenesssomeone missing · likeness weak
Seedream 5.0 ProOriginal
Passed every check
FLUX 3Original
Usable · partly: likenesslikeness weak
GPT Image 2.5Original
Usable · partly: likenesslikeness weak
GPT Image 2Original
Passed every check
Seedream 5.0 FlashOriginal
Passed every check
Three people on a hike
All eight are correct; the GPT models and Seedream 5.0 Flash kept faces best.
The man from the first photo, the man from the second photo and the woman from the third photo on a mountain trail with backpacks, peaks behind them, bright daylight, smiling at the camera, realistic photo.
Nano Banana 2.1Original
Passed every check
Nano Banana 2Original
Usable · partly: likenesslikeness weak
Nano Banana ProOriginal
Usable · partly: likenesslikeness weak
Seedream 5.0 ProOriginal
Usable · partly: likenesslikeness weak
FLUX 3Original
Usable · partly: likenesslikeness weak
GPT Image 2.5Original
Passed every check
GPT Image 2Original
Passed every check
Seedream 5.0 FlashOriginal
Passed every check
Adding a missing person to a group photo
We gave each model a photo of three friends at dinner and a photo of a fourth person, and asked for him to join them. Six models added him and kept the scene. Nano Banana Pro and Seedream 5.0 Pro replaced one of the three with him; GPT Image 2 added him but rearranged the table.
Add the man from the second photo to the group in the first photo, sitting at the table with them and raising a glass. Keep the scene and the three people already there the same, realistic photo.
Nano Banana 2.1Original
Passed every check
Nano Banana 2Original
Passed every check
Nano Banana ProOriginal
Not usable · failed: everyone there · partly: likeness, as describedsomeone missing · likeness weak
Seedream 5.0 ProOriginal
Not usable · failed: everyone there · partly: likeness, as describedsomeone missing · likeness weak
FLUX 3Original
Passed every check
GPT Image 2.5Original
Passed every check
GPT Image 2Original
Usable · partly: as describedframing changed
Seedream 5.0 FlashOriginal
Passed every check
The same request, three times
Every photo above is one try. To see how steady a model is, we sent two of these descriptions to PicYep's default model, Nano Banana 2.1, two more times each, with the same photos and settings, and checked every result the same way (October 15, 2026). Five of the six tries were usable. The holiday party passed every check all three times. The patio dinner kept the faces only partly each time (0.73 to 0.75), and its third try added a fourth person at the table, so it failed. In every case the first try was already usable.
| # | Try | Usable | Passed every check | Face match | Credits |
|---|---|---|---|---|---|
| 1 | Try 1 | 2 of 2 | 1 of 2 | 0.78 | 30 |
| 2 | Try 2 | 2 of 2 | 1 of 2 | 0.77 | 30 |
| 3 | Try 3 | 1 of 2 | 1 of 2 | 0.79 | 30 |
The woman from the first photo, the man from the second photo and the woman from the third photo, friends having dinner together at a long wooden table on a patio, string lights overhead, laughing and raising their glasses, warm evening light, a candid photo, realistic skin texture.
The four people from the four photos at a holiday party in a cozy living room, standing together in front of a decorated Christmas tree, holding mugs of cocoa, smiling at the camera, warm lights, realistic photo.
Eight, twelve and sixteen people
The tests above stop at six people. To see what happens with bigger groups, we added more AI-generated people, sixteen in all and each easy to tell apart (FaceNet scores between them run from −0.29 to 0.53), and asked for a barbecue of eight, a team photo of twelve and a family reunion of sixteen, wide (3:2) at 2K (October 22, 2026). Each description is written the way PicYep's Assistant writes one for six or more people: how many people there are, that each appears once, and a numbered line for each photo.
We used the two models PicYep sends big groups to. Nano Banana 2.1 reads up to 14 photos; we asked it three times for eight and for twelve people. GPT Image 2.5 reads up to 16; we asked it once for eight and for twelve, and three times for sixteen. We counted every person by eye and measured every face.
- Eight people: all four photos usable, everyone in each, face match 0.72 to 0.79.
- Twelve people: Nano Banana 2.1 made one of three usable. One try kept twelve people but turned two of them into strangers; another kept everyone, with faces too loose to pass (0.50). GPT Image 2.5 kept all twelve (0.77).
- Sixteen people: GPT Image 2.5 kept all sixteen in two of three tries (0.75 and 0.74). The third left two people out and merged two men into one face.
No photo in this round passed every check: in a wide photo of eight people or more each face is small, and none reached our 0.80 likeness mark. The more people, the more chance that someone is left out, which is why PicYep checks a big group photo against its list.
Nano Banana 2.1
| # | Try | Usable | Passed every check | Face match | Credits |
|---|---|---|---|---|---|
| 1 | Try 1 | 1 of 2 | 0 of 2 | 0.57 | 48 |
| 2 | Try 2 | 2 of 2 | 0 of 2 | 0.72 | 48 |
| 3 | Try 3 | 1 of 2 | 0 of 2 | 0.62 | 48 |
Eight friends at a backyard barbecue on a summer evening, gathered around a wooden picnic table with string lights overhead, laughing and smiling at the camera. The photo has exactly eight people, each appearing exactly once, with no one added and no one left out. Keep each person's face, hair, glasses, beard and clothing from their reference photo. Realistic photograph, wide framing, every face clearly visible. The reference images, in order: image 1: a woman with long straight black hair, grey T-shirt; image 2: a man with a short grey beard, navy polo shirt; image 3: an older woman with short grey hair and round glasses, green cardigan; image 4: a young man with dark wavy hair and light stubble, white T-shirt; image 5: a man with short curly hair and a dark mustache, olive T-shirt; image 6: a woman with shoulder-length black braids and gold hoop earrings, mustard T-shirt; image 7: a man with short dark hair and black-rimmed glasses, grey hoodie; image 8: an older man with short greying hair and a grey beard, blue sweater.
Try 1
Usable · partly: likenesslikeness weak
Try 2
Usable · partly: likenesslikeness weak
Try 3
Usable · partly: likenesslikeness weak
A team photo of twelve coworkers outdoors in front of a modern glass office building, standing in two rows, smart-casual clothes, bright daylight, smiling at the camera. The photo has exactly twelve people, each appearing exactly once, with no one added and no one left out. Keep each person's face, hair, glasses, beard and clothing from their reference photo. Realistic photograph, wide framing, every face clearly visible. The reference images, in order: image 1: a woman with long straight black hair, grey T-shirt; image 2: a man with a short grey beard, navy polo shirt; image 3: an older woman with short grey hair and round glasses, green cardigan; image 4: a young man with dark wavy hair and light stubble, white T-shirt; image 5: a man with short curly hair and a dark mustache, olive T-shirt; image 6: a woman with shoulder-length black braids and gold hoop earrings, mustard T-shirt; image 7: a man with short dark hair and black-rimmed glasses, grey hoodie; image 8: an older man with short greying hair and a grey beard, blue sweater; image 9: a woman with long dark brown hair, black top; image 10: a man with very short hair, dark crew-neck sweater; image 11: an older East Asian man with a shaved head and a grey goatee, burgundy sweater; image 12: a young woman with curly copper-red hair and freckles, denim shirt.
Try 1
Not usable · failed: everyone there, likenessface changed · likeness weak
Try 2
Usable · partly: likenesslikeness weak
Try 3
Not usable · failed: likenesslikeness weak
GPT Image 2.5
| # | Try | Usable | Passed every check | Face match | Credits |
|---|---|---|---|---|---|
| 1 | Try 1 | 3 of 3 | 0 of 3 | 0.77 | 135 |
| 2 | Try 2 | 0 of 3 | 0 of 3 | 0.64 | 165 |
| 3 | Try 3 | 1 of 3 | 0 of 3 | 0.74 | 165 |
Eight friends at a backyard barbecue on a summer evening, gathered around a wooden picnic table with string lights overhead, laughing and smiling at the camera. The photo has exactly eight people, each appearing exactly once, with no one added and no one left out. Keep each person's face, hair, glasses, beard and clothing from their reference photo. Realistic photograph, wide framing, every face clearly visible. The reference images, in order: image 1: a woman with long straight black hair, grey T-shirt; image 2: a man with a short grey beard, navy polo shirt; image 3: an older woman with short grey hair and round glasses, green cardigan; image 4: a young man with dark wavy hair and light stubble, white T-shirt; image 5: a man with short curly hair and a dark mustache, olive T-shirt; image 6: a woman with shoulder-length black braids and gold hoop earrings, mustard T-shirt; image 7: a man with short dark hair and black-rimmed glasses, grey hoodie; image 8: an older man with short greying hair and a grey beard, blue sweater.
Try 1
Usable · partly: likenesslikeness weak
A team photo of twelve coworkers outdoors in front of a modern glass office building, standing in two rows, smart-casual clothes, bright daylight, smiling at the camera. The photo has exactly twelve people, each appearing exactly once, with no one added and no one left out. Keep each person's face, hair, glasses, beard and clothing from their reference photo. Realistic photograph, wide framing, every face clearly visible. The reference images, in order: image 1: a woman with long straight black hair, grey T-shirt; image 2: a man with a short grey beard, navy polo shirt; image 3: an older woman with short grey hair and round glasses, green cardigan; image 4: a young man with dark wavy hair and light stubble, white T-shirt; image 5: a man with short curly hair and a dark mustache, olive T-shirt; image 6: a woman with shoulder-length black braids and gold hoop earrings, mustard T-shirt; image 7: a man with short dark hair and black-rimmed glasses, grey hoodie; image 8: an older man with short greying hair and a grey beard, blue sweater; image 9: a woman with long dark brown hair, black top; image 10: a man with very short hair, dark crew-neck sweater; image 11: an older East Asian man with a shaved head and a grey goatee, burgundy sweater; image 12: a young woman with curly copper-red hair and freckles, denim shirt.
Try 1
Usable · partly: likenesslikeness weak
A family reunion of sixteen people in a sunny garden, three generations standing and sitting together in three rows on and around a long bench, the baby held in the arms of an adult in the front row, everyone smiling at the camera. The photo has exactly sixteen people, each appearing exactly once, with no one added and no one left out. Keep each person's face, hair, glasses, beard and clothing from their reference photo. Realistic photograph, wide framing, every face clearly visible. The reference images, in order: image 1: a woman with long straight black hair, grey T-shirt; image 2: a man with a short grey beard, navy polo shirt; image 3: an older woman with short grey hair and round glasses, green cardigan; image 4: a young man with dark wavy hair and light stubble, white T-shirt; image 5: a man with short curly hair and a dark mustache, olive T-shirt; image 6: a woman with shoulder-length black braids and gold hoop earrings, mustard T-shirt; image 7: a man with short dark hair and black-rimmed glasses, grey hoodie; image 8: an older man with short greying hair and a grey beard, blue sweater; image 9: a woman with long dark brown hair, black top; image 10: a man with very short hair, dark crew-neck sweater; image 11: an older East Asian man with a shaved head and a grey goatee, burgundy sweater; image 12: a young woman with curly copper-red hair and freckles, denim shirt; image 13: an older woman with silver hair in a low bun and gold earrings, teal top; image 14: a teenage boy with short twists, yellow hoodie; image 15: a blond man with a light-brown beard, red flannel shirt; image 16: a baby in a white bodysuit.
Try 1
Usable · partly: likenesslikeness weak
Try 2
Not usable · failed: everyone there · partly: likenesssomeone missing · likeness weak
Try 3
Usable · partly: likenesslikeness weak
How PicYep handles six people or more
When you add six photos or more in the Assistant, it lists them with a number and a short description of each, for you to check before anything is made, and suggests a wide shape at 2K. The image model is told who is who in that order. When the photo is ready, it is checked against the list: if someone is missing or appears twice, the Assistant says who, shows their photo, and offers to add only them or to make the photo again for free (one free remake a day). The check reads faces too, so a person who looks very like another can slip past it; look over a big group photo yourself before you share it.
Model by model
- Seedream 5.0 Flash (15 credits) — 6 of 6 usable, 5 passed every check. Everyone in every photo, faces among the closest, and the lowest price.
- GPT Image 2 (50–75 credits for a group, medium quality) — 6 of 6 usable, 5 passed every check. The highest face match; it rearranged the scene when adding a person.
- GPT Image 2.5 (25–50 credits for a group, medium quality) — 6 of 6 usable, 4 passed every check. Faces almost as close as GPT Image 2; it cropped the team photo too tightly.
- Nano Banana 2.1 (25–35 credits) — 5 of 6 usable, 4 passed every check. Everyone kept and natural scenes, but looser likenesses; the six-person picnic fell just under our likeness threshold.
- Nano Banana 2 (40 credits) — 6 of 6 usable, 2 passed every check. Everyone kept; likeness looser in four of six photos.
- FLUX 3 (25 credits) — 6 of 6 usable, 2 passed every check. Everyone kept; looser likenesses and one scene not as described.
- Seedream 5.0 Pro (35 credits) — 5 of 6 usable, 2 passed every check. Swapped a person out when asked to add one.
- Nano Banana Pro (75 credits) — 4 of 6 usable, none passed every check. Left a person out of two photos and had the loosest likenesses — the most expensive model did worst here.
Photo editing or AI generation?
When asked how to combine people from separate photos, ChatGPT pointed to Photoshop's Generative Fill (and its own image tool, and Canva). They work differently:
- Photoshop Generative Fill adds or removes content in a selected area from a text prompt, in Photoshop on desktop, web and mobile, with a membership or free trial and a set number of Generative Credits (Adobe's page, checked October 15, 2026). Its page does not say it combines people from separate photos; you place and blend each person yourself.
- An AI group photo generator draws one new photo with everyone in it, in the same light and setting, from a photo of each person — no cutting out or blending.
If you need every pixel of an existing photo kept, edit it. If you need everyone together in a scene that never happened, generate it.
For prices, refunds and privacy across more AI photo services, see our comparison.
Which one should you use?
- Most groups: Seedream 5.0 Flash — everyone kept, faces close, 15 credits.
- Faces matter most: GPT Image 2, or GPT Image 2.5 for less.
- Adding someone to an existing photo: Seedream 5.0 Flash, Nano Banana 2.1, GPT Image 2.5 or FLUX 3 — avoid Nano Banana Pro and Seedream 5.0 Pro for this.
- Five or six people: pick a wider frame (4:3 or 16:9) so nobody is squeezed, and expect slightly looser faces in wide candid shots.
- Eight people or more: a wide shape at 2K, and a look over the result for anyone missing; PicYep's Assistant suggests the first and checks the second for you.
A description to copy
The [man/woman] from the first photo, the [man/woman] from the second photo and the [man/woman] from the third photo, [what they are doing] in [where], smiling at the camera, realistic photo.
To add someone to a photo you have: Add the [man/woman] from the second photo to the group in the first photo, [where and how]. Keep the scene and the people already there the same, realistic photo.
Questions
Can AI put people from different photos into one picture?
Yes. In our test, eight AI models each made six group photos from separate photos of three to six people. Six of the eight put every person in every photo; Nano Banana Pro left someone out twice and Seedream 5.0 Pro once.
Which AI is best for making a group photo from separate photos?
Seedream 5.0 Flash and GPT Image 2 both made all six photos usable, five of them passing every check, and kept faces closest to the input photos. Seedream 5.0 Flash costs 15 credits a photo; GPT Image 2 costs 50 to 75 credits for a group, depending on how many photos you add.
How many people can be in one AI group photo?
PicYep takes up to 16 photos in one request, as many as the chosen model reads. With three to six people, every model but Nano Banana Pro kept everyone. With more, we tested eight, twelve and sixteen: all eight were kept every time; twelve were kept by GPT Image 2.5 and in two of three tries by Nano Banana 2.1; sixteen were kept in two of three tries by GPT Image 2.5. For six people or more, PicYep checks the photo against its list and offers to add whoever is missing.
Can AI add a missing person to a group photo?
Yes. We added a fourth person to an existing photo of three friends at dinner. Six of the eight models added him and kept the other three; Nano Banana Pro and Seedream 5.0 Pro swapped one of the original three out instead.
Will everyone still look like themselves?
Mostly. We measured every face with FaceNet against its input photo: GPT Image 2, GPT Image 2.5 and Seedream 5.0 Flash averaged 0.86 to 0.88, the other models 0.69 to 0.79 (1.00 is the same face; different people in our set score 0.22 or lower). Faces that are small or turned in a wide candid shot measure lower.
How do I tell the AI who is who?
Add one photo per person and name them by order in your description: "the woman from the first photo, the man from the second photo…". For four or more people, "the five people from the five photos" also worked in our test.
