The gallery holds the six published examples from the model page, copied here so the review keeps them, plus one clip of our own from 8 October that pushes the reference count to five. Everything in it shows clothed adults, suggestive at most. We do not publish our content-ladder runs.
A line to camera (image to video, 1080p). From a frame of a woman with dark wavy hair in a black satin slip dress at a three-way mirror under vanity bulbs, the dress unzipped a few inches at the back. She turns to the camera, lets the loose strap slide further, reaches for the zip and says "Help me with this?", then bites her lip. Her face, the dress and the room stay as the frame drew them, and the 16:9 frame came back at exactly 1920 x 1080.
A walk out of the steam (image to video, vertical). A man with stubble in a white towel rises from the bench of a marble bath, pushes his wet hair back and walks toward the camera. The 9:16 frame came back at 720 x 1280, and the sound carries the hiss of steam and dripping water.
A morning kiss (image to video). Two women on a rumpled bed in morning light, one with a platinum crop in a champagne silk camisole, the other with long auburn hair in an unbuttoned white shirt, kiss slowly and smile.
Three references, one fitting (reference to video, 16:9). A portrait of a woman with short natural curls and gold hoops, a product photo of an emerald silk robe on a hanger and an empty boutique fitting room. She wears the robe over a black slip in that room, ties the sash, turns to the gilded mirror and lets the robe slide off one shoulder, glancing back at the camera.
A line in a borrowed voice (reference to video, 9:16, 9 seconds). One portrait of a woman with long black hair and red lipstick plus one voice clip. In a candlelit doorway she turns the key and whispers "Lock the door. Nobody is coming up tonight... and I am not in any hurry." The line is written in the prompt; the audio clip only sets the voice.
Two portraits, one elevator (reference to video, 9:16). Two portraits of two men in suits. In a mirrored brass hotel elevator the bearded one loosens the other's tie and backs him against the mirror, and they almost kiss as the doors open. Both faces are kept.
Five references, three people (reference to video, 16:9, our own test). We sent two portraits of women, the bearded man's portrait, the emerald robe and the fitting room, and asked for the woman with long black hair to try on the robe while the woman with short curls ties the sash and the man says "It's perfect. We're late." from the doorway. All three faces came through, and so did the room: the pink velvet curtains, the gilded mirror and the rail of slips. The line was spoken (our transcript read "It's perfect, we're late."). Two things drifted: the robe ended up on both women (the prompt never said what the second woman wore), and the man stood by the rail rather than leaning in the doorway. The clip took about 8 minutes to come back.
Sizes, files and speed
Image to video follows your frame. One 16:9 frame came back at 952 x 544 at 540p, 1268 x 724 at 720p, 1906 x 1088 at 1080p and 3804 x 2172 at 4k; another came back at exactly 1920 x 1080 at 1080p; a 2:3 portrait gave 586 x 880 at 540p. On reference to video, 16:9 at 540p gave 960 x 528, 9:16 at 720p gave 720 x 1280 and 1:1 at 2k gave 1920 x 1920. Clips run a few hundredths of a second over the requested length (5 seconds came back as 5.04).
A sensible loop: draft at 540p for 5 seconds, which took about 2 minutes on image to video, and rerun the take you like at 1080p. Choose 2k or 4k only when you need the size and your editor handles HEVC. Budget 4 to 6 minutes per reference-to-video clip, and write the whole scene before you send it, because a second try costs another wait.
