A painting filter adds texture to a photo and leaves everything else where it was, which is why the result still looks like a photo. A classical oil painting couple portrait is made of decisions the camera did not take: one light instead of several, shadows that hold no detail, and a brush that shows. This article converts one photo of a couple into such a painting in CapCut at 16:9, measures what the conversion changed and what it kept, and comes back with a specific answer about the light that decides which photo you should feed it in the first place.
The photo is generated too, on purpose. No real couple's picture is used here, so the first step is to make a photo with the problems a phone photo has, and only then to paint it. Everything below happened in Video Studio's Image mode on 5 September 2026, five images at one credit each.
Make the photo first, and make it ordinary
The photo has to be ordinary in a particular way: lit from two sides so that nothing on the faces is in shadow, sharp everywhere, with a living room behind it. That is the picture most people would hand to a painter, and it is the hardest case for the painting, so it is the one to test.
Photoreal snapshot, 16:9 landscape, taken indoors on a phone with flat, even light: a ceiling light and a window on the left both lighting the scene so there are almost no shadows on the faces. A couple in their early thirties in a plain living room: a woman with shoulder-length dark brown hair in a mustard-yellow sweater stands slightly in front, and a man with short black hair, a light stubble and a gray crewneck stands just behind her with his arm around her shoulders and his cheek near her temple, both looking at the camera and smiling gently. A beige sofa and a bookshelf out of focus behind them, ordinary colors, sharp detail everywhere, no film grain, no filter. The couple occupies the center and right of the frame with room on the left. No text, no logos, no brand marks.
It came back through the CapCut AI image generator flow at Seedream 4.5, 16:9 and 2K, 2560 by 1440 pixels. Two numbers describe how flat it is. Splitting each face down the middle and averaging the brightness of each half, on a 0 to 255 scale, gives 121 against 96 for her and 87 against 93 for him: the window puts a little more light on her left, and his face is lit evenly. Only 5 percent of the frame is darker than 40. A painter would call this a photo with no dark in it.
Ask for the oil painting couple portrait and look at what it keeps
The photo was dropped onto the home composer, the mode menu was set to Image, and the settings row read Seedream 4.5, 16:9, 2K. The prompt asks for the three things a painting is made of, and it asks to keep the people.
Turn this photo into a classical oil painting portrait of the same two people, keeping their faces, hair, expressions and the way they stand exactly as they are. One strong light only, coming from the upper left, so the left side of each face is lit and the right side falls into deep shadow; the shadows are painted dark and simple with no detail in them. The background is a plain dark brown-black, like an old studio portrait, with no sofa and no bookshelf. Visible brushstrokes everywhere: thick paint in the highlights on the cheeks, nose and knuckles, softer blended strokes in the shadows, loose strokes in the clothing, and the sweater and crewneck kept in their colors but painted, not photographed. Muted, slightly warm palette, no glossy skin, no photographic sharpness, no lens blur. 16:9 landscape, same framing as the photo. No text, no signature, no frame.
The composer in Image mode with the photo attached. The same row has to read 16:9, or the painting comes back in whatever shape the photo suggests.
The photo and the first conversion. The room is gone, the sweater is paint, the faces are the same faces, and the light on them is the same light.
What it kept is the good news: the same two faces, the same expressions, the same hand on the shoulder, the same mustard and gray. The living room became a dark brown ground with a brush in it, and the share of the frame darker than 40 went from 5 percent to 68 percent. What it also kept is the problem. Her face measures 126 against 91 and his 83 against 73, a shade more directional than the photo and nowhere near the "deep shadow" the prompt asked for. The dark in this painting is all in the background; the people are still lit by the ceiling light that is no longer in the picture. In this run, the conversion took the photo's light as part of the photo.
The cost line before the photo and before the conversion, both 1 credit at Image mode, Seedream 4.5, 16:9 and 2K on 5 September 2026.
Each image in this article was preceded by the line "Generating 1 item will consume 1 credit" in the chat, at Image mode with Seedream 4.5, 16:9 and 2K. Other models, resolutions and modes are priced differently, and video is priced separately, so read the line before you send rather than carrying this number over. The chat's reply after the conversion also offered three chips, "Animate this image", "Generate 4 variations" and "Make the lighting warmer", and the card's toolbar on the canvas carried "Image to video" and "Image to image" buttons.
Ask again with the light first
The obvious repair is to put the lighting at the front of the request and to stop asking for the faces "exactly as they are", since that phrase may be holding the light in place as well. The second conversion used the same photo and this prompt.
Repaint this photo as a classical oil painting and change the lighting. There is one strong light only, from the upper left: the left side of each face is lit, and the right half of each face, the right side of the necks and the right side of both bodies fall into deep shadow, painted dark and simple with no detail in the shadow. Keep the same two people, the same faces, the same expressions, the same pose and the same framing. Plain dark brown-black background like an old studio portrait, no sofa, no bookshelf. Visible brushstrokes: thick paint in the lit highlights on the cheeks, nose and knuckles, soft blended strokes in the shadows, loose strokes in the clothing, the sweater and crewneck painted in their colors. Muted, slightly warm palette, no glossy skin, no photographic sharpness. 16:9 landscape. No text, no signature, no frame.
The second conversion. The wall now shows where the light comes from; his face moved toward it, hers barely did.
Half of it worked. A beam of light appeared on the wall from the upper left, exactly where the prompt put the source, and his face went from 87 against 93 in the photo to 62 against 39, a real shadow side. Her face went from 121 against 96 to 140 against 95: brighter on the lit side, no darker on the other. The wording moved the person who was already half in shadow behind her and left the person in front lit the way the photo lit her. Two conversions of the same photo, one with the light asked for politely and one with it demanded, and the front face kept the photo's light both times.
Take the photo for the light you want in the oil painting
If the conversion keeps the photo's light, the light has to be in the photo. The third source image is the same couple and the same room with the ceiling light off and only the window on the left, and it was painted with the first prompt, the one that did not manage to relight the flat photo.
Photoreal snapshot, 16:9 landscape, taken indoors on a phone in the late afternoon with the ceiling light off: the only light is a window on the left, so the left side of each face is bright and the right side falls into soft shadow. A couple in their early thirties in a plain living room: a woman with shoulder-length dark brown hair in a mustard-yellow sweater stands slightly in front, and a man with short black hair, a light stubble and a gray crewneck stands just behind her with his arm around her shoulders and his cheek near her temple, both looking at the camera and smiling gently. A beige sofa and a bookshelf out of focus behind them in the dimmer part of the room, ordinary colors, sharp detail, no film grain, no filter. The couple occupies the center and right of the frame with room on the left. No text, no logos, no brand marks.
The side-lit photo and the finished painting. The faces keep the photo's light ratio almost exactly; the painting adds the ground, the beam and the brush.
This is the finished portrait. In the photo the faces measure 97 against 75 for her and 60 against 44 for him, and in the painting 127 against 98 and 81 against 64: the painting lifted both faces by about a third and kept the ratio between the two halves, 1.3 for her in both, 1.35 and 1.25 for him. The share of the frame darker than 40 is 26 percent in the photo and 67 percent in the painting, and this time the dark reaches the people, because the photo brought it in. The man in the new photo looks down at her rather than at the camera, which the prompt did not ask for and which suits a portrait better than the pose it did ask for; that is a generated photo's freedom, not a rule about real ones.
The practical reading is short. Choose or shoot the photo for the painting's light: one window, the room lights off, the couple turned so the window is to one side. The conversion will keep that light in this run's manner, and the prompt's lighting sentence then describes what is already there instead of arguing with it.
Show the brush where a brush can be seen
Where the brush shows: the sweater and the background. At this size the strokes on the sweater are ridges of paint; the wall is dragged brush over a dark ground.
The brush lives in the clothing and the ground, not in the faces. On the sweater the knit became ridges of thick paint following the folds, and the knuckles on her shoulder got the impasto highlights the prompt asked for. The background became a dragged, uneven dark with the beam of light brushed across it. The faces, at the same magnification, are painted more cautiously: skin texture and softened edges, but the eyes, the smile and the hairline read as they did in the photo. Since likeness is the thing a couple will check first, this is the right place for the brush to be timid, and the prompt's "keeping their faces" did its work. Two marks appeared that the prompt tried to exclude: a faint scribble like a signature in the lower right corner of the finished painting, and a small "Ai" label in the top-left corner of the images downloaded in this session. Labeling and provenance indicators may vary by workflow and current CapCut implementation, so check the final downloaded file before printing.
How big it prints, and whose faces they are
The painting is 2560 by 1440 pixels, the 2K size the settings row promised. At 300 pixels per inch, the density print shops usually ask for, that is 8.5 by 4.8 inches, about 22 by 12 centimeters; at 150 pixels per inch, 17 by 9.6 inches. It is a wide picture, since the whole series is made at 16:9, so it frames as a panel rather than a portrait-shaped canvas; the empty left third with the beam on it is part of the composition, not waste to crop off. No larger export was tested here; 2K is the size the settings row showed for this run, and nothing above it is claimed.
The two people in this article are not real, which is the only reason their faces are in it. With a real photo, the faces belong to the people in it: paint your own photo, or one whose subjects have agreed to be painted, and treat the finished image the way you would treat the photo, since it keeps their likeness by design. The next step is the one the third section earned: take the photo with the window to one side and the lights off, drop it on the composer, and send the first prompt.
Written 5 September 2026. Both source photos and all three paintings are generated images from CapCut made for this article; no real person, product or brand appears. Interface labels, chip texts, the cost line and pixel sizes reflect a single session on that date and the configuration named above, and may change. Brightness figures are averages of the left and right halves of a face box on the downloaded file, on a 0 to 255 scale. Background removal, printing services and framing are not covered.