> ## Content Index > Fetch the complete content index at: https://theseguysknow.io/llms.txt > Use this file to discover other available public pages before exploring further. # Best AI Image-to-Video Generators 2026: 6 Models Tested - URL: https://theseguysknow.io/best-ai-image-to-video-generators/ - Published: 2026-08-12T15:54:44.000Z - Updated: 2026-11-10T08:28:27.000Z - Description: The same three source images, run through every leading image-to-video model, with each first take, prompt and price on the page. - Author: Mike Hazard - Tags: AI Tools & Models, #how we know, #tgk-pick:apps-tools:image-to-video:per_use:Kling O3 Pro (about $0.56 per 5s clip) Kling 4.0 lost to the older Kling O3 Pro, and a $0.50 Wan 3.0 clip beat both of them on a close-up face. That is what you find out when you stop watching launch demos and pay for the clips yourself. We gave six image-to-video models the same three Nano Banana Pro images, a close portrait, a poolside walk and a perfume ad where her hand lifts a glass bottle, and kept each model's first result. Seedance 2.5, Kling O3 Pro, Veo 3.1 and Wan 2.7 Pro ran in August, and Kling 4.0 and Wan 3.0 joined in October. ## Quick Answer Seedance 2.5 still made the best videos, with faces that looked the most human, eyes and small expressions none of the others matched, and the cleanest handling of the perfume bottle. When quality matters more than price, we pay for Seedance. Wan 3.0 was the surprise of the October update. It beat both Kling models on the portrait, made the most interesting perfume clip and costs $0.50 per clip at 720p, which makes it our value pick and second overall. Kling O3 Pro came third and still looks like a bargain at $0.56\. Kling 4.0 finished fourth because the portrait smile went wrong and it costs more than its older brother. Veo 3.1 was fifth, and Wan 2.7 Pro was clearly last for this kind of human, fashion and advertising work. | Rank | Model | Best part of my test | Main weakness | Approx. cost per test | | ---- | -------------- | --------------------------------------------------------------------------- | ------------------------------------------------------------ | --------------------- | | 1 | Seedance 2.5 | Faces, micro-expressions and product interaction | Expensive; hair can still move like one solid piece | $1.80 | | 2 | Wan 3.0 | Natural portrait smile, visible pores and the most interesting perfume lift | Hair moves in chunks; faint cartoon outlines on the legs | $0.50 | | 3 | Kling O3 Pro | Walking, prompt obedience and value | Slight CGI layer on skin | $0.56 | | 4 | Kling 4.0 | Realistic fine hair and a natural perfume pose | Wide open-mouth smile where a subtle one was asked for | $0.79 | | 5 | Google Veo 3.1 | Cinematic framing and some good hair movement | Artificial facial texture and a visible edge around movement | $1.20 | | 6 | Wan 2.7 Pro | Can produce interesting experimental motion | Face drift, harsh skin and weaker object consistency | From $0.60 | Prices are WaveSpeed list prices before account discounts, checked on August 12, 2026 for the first four models and October 26, 2026 for Kling 4.0 and Wan 3.0\. My discounts brought the October clips down to $0.59 for Kling 4.0 and $0.475 for Wan 3.0\. Wan 2.7 Pro was listed from $0.60, and we did not record its exact charge. ## How We Tested the Six Models All 18 videos ran through WaveSpeed, so every model went through the same workflow and the same bill. This covers image-to-video only, with no text-to-video, editing, reference-to-video or long scenes. The rules were simple: - Three 4K source images from Nano Banana Pro, the same files for every model. - One generation per image and model, with no rerolls or replacement clips. - 16:9 output and prompt enhancer off. - No last frame or extra character references. - Each clip judged on motion, identity, hands, objects, image quality and whether we would use it. Each model reads prompts differently, so every model got the same action and camera direction in the format it handles best. Seedance got a chronological description, both Klings got separate subject, camera and preservation blocks, Veo got conventional shot direction and both Wans got a short command. Kling 4.0 and Wan 3.0 used the same prompts as their older versions, word for word, and every prompt is on the page so you can judge them yourself. | Model endpoint on WaveSpeed | Duration | Output used | Audio | | ----------------------------------------- | -------- | ----------- | ----- | | bytedance/seedance-2.5/image-to-video | 5 sec | 720p | On | | alibaba/wan-3.0/image-to-video | 5 sec | 720p | On | | kwaivgi/kling-video-o3-pro/image-to-video | 5 sec | 1080p | Off | | kwaivgi/kling-v4.0/image-to-video | 5 sec | 720p | On | | google/veo3.1/image-to-video | 6 sec | 1080p | Off | | alibaba/wan-2.7/image-to-video-pro | 5 sec | 1080p | On | Veo only offers four, six or eight seconds, so it got six, because four rushed the perfume lift. WaveSpeed charged the same for Veo at 720p and 1080p without sound, so we took 1080p, and Wan 2.7 Pro’s endpoint starts at 1080p anyway. Seedance stayed at 720p because 1080p would have raised each clip from $1.80 to $4.50\. Kling 4.0 and Wan 3.0 ran at 720p with audio on. [Google’s own image-to-video guidance](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/video/best-practice?ref=theseguysknow.io) says the source image sets the character, lighting and style, so a sharp starting image matters, and this test backed that up. At list prices, the videos came to at least $16.34 for all 18 clips, counting Wan 2.7 Pro at its starting price. After account discounts, the six October clips cost us $3.19. | Model | Three tests | | -------------- | ----------- | | Seedance 2.5 | $5.40 | | Wan 3.0 | $1.50 | | Kling O3 Pro | $1.68 | | Kling 4.0 | $2.36 | | Google Veo 3.1 | $3.60 | | Wan 2.7 Pro | From $1.80 | The total leaves out the source images, which started at $0.14 each for [Nano Banana Pro on WaveSpeed](https://wavespeed.ai/models/google/nano-banana-pro/text-to-image?ref=theseguysknow.io), because picking them can take rerolls and that cost belongs to a different test. ## Test 1: A Close Portrait Exposed the Skin and Eyes The first source was a simple close portrait of a Mediterranean woman looking straight at the camera on a rooftop at golden hour, and the animation only asked for a blink, a small smile, a slight head turn, moving hair and a slow push-in. That simplicity is the point, because a close-up gives the model nowhere to hide. If the pupil slides, the skin gains a strange digital pattern or the hair moves like a detachable wig, you are looking straight at it. ![Close portrait of a beautiful 25-year-old Mediterranean woman standing on a bright rooftop terrace at golden hour.](https://storage.ghost.io/c/21/2f/212fa2d6-dfa3-4908-9426-ec8fc852e19d/content/images/2026/08/TEST_SOURCE_1.jpg) **The untouched 4K Nano Banana Pro source used for all six portrait videos.* #### See the portrait source-image prompt and settings ****Model:** Nano Banana Pro ****Settings:** 16:9, 4K, highest clean quality available ****Prompt:** `Photorealistic cinematic close portrait of a beautiful 25-year-old Mediterranean woman standing on a bright rooftop terrace at golden hour. Long dark brown hair moving slightly in the breeze, warm olive skin, expressive brown eyes, natural makeup, small gold earrings and an elegant cream silk blouse. She looks directly into the camera with a relaxed, slightly playful expression. Soft sunlight across one side of her face, realistic skin texture, shallow depth of field, premium fashion editorial photography, natural proportions, clean uncluttered background. Horizontal 16:9 composition.` #### Seedance 2.5 Had the Best Eyes Seedance won this test with a natural blink, a face that stayed recognisable and eyes that kept the small reflections and colour variations that make them feel alive, while the skin stayed close to the source without turning waxy or oversharpened. Its hair was among the better results too, though that is a low bar, because every current model still locks separate strands together and moves them as one heavy piece, which gives hair that familiar wig-like movement. We have already tested Seedance on much harder movement and a 30-second tennis scene in our [full Seedance 2.5 review](https://theseguysknow.io/seedance-2-5-review/), so this portrait was an easier job, and it handled it almost exactly as we hoped. 0:00 /0:05 1× **Seedance kept the eye detail, skin and identity closest to the source.* #### See the Seedance portrait prompt and settings ****Settings:** 5 sec, 720p, 16:9, audio on, no last image, prompt enhancer off ****Prompt:** `5-second continuous fashion portrait shot. The woman holds relaxed eye contact with the camera. She blinks naturally once, then her expression gradually develops into a subtle warm smile as she turns her face only a few degrees toward the sunlight. A gentle rooftop breeze moves a few loose strands of hair across her shoulder. The camera performs a very slow, smooth push-in. Preserve her exact facial features, age, hairstyle, clothing, jewellery, body position, rooftop setting and golden-hour lighting throughout. Restrained realistic motion with no scene change or cut.` ##### Wan 3.0 Beat Both Klings on the Portrait Wan 3.0 framed her closer than the other models and then pushed in again, which helped it, and the result had a good smile, visible pores and no strange artifacts on the face that we could find. Where it slipped was the hair, which moved in chunks with different sections living their own lives, a problem every model on this page still has. Next to its little brother Wan 2.7 Pro this is a different league, and on the face alone it beat both Kling models. 0:00 /0:05 1× Wan 3.0 kept a natural smile and visible skin pores, although the hair moved in chunks. #### See the Wan 3.0 portrait prompt and settings Settings: 5 sec, 720p, 16:9, audio on, prompt expansion off, no last image Prompt: The woman looks directly into the camera. She blinks once naturally, gives a small warm smile, and turns her head slightly toward the sunlight. A gentle breeze moves a few strands of her hair. Slow camera push-in. One continuous realistic fashion portrait shot. Keep her facial appearance, hair, cream blouse, jewellery, pose, lighting and rooftop background consistent. #### Kling O3 Pro Was Close, Especially for $0.56 Kling O3 Pro also gave us a good blink, sensible eye movement and a decent push-in, and its gentler wind kept the hair under control. Watched straight after Seedance, though, its eyes had less depth and the skin looked as though the model had laid a thin CGI layer over the original face to keep it under control. We cannot say that is technically what Kling does, but it is how the finished video looked to us. Even so, at $0.56 this was a very good result that never looked like the cheap version of the test. 0:00 /0:05 1× **Kling stayed natural, but the skin and eyes lost some of the detail Seedance preserved.* #### See the Kling portrait prompt and settings ****Settings:** 5 sec, 1080p, 16:9, audio off, one starting image, prompt enhancer off ****Prompt:** SUBJECT ACTION: The woman maintains direct eye contact, blinks naturally once and slowly forms a subtle warm smile. She turns her face slightly toward the sunlight. Only a few loose strands of hair move in the gentle breeze. CAMERA: Very slow cinematic dolly-in. One continuous shot. `PRESERVATION: Keep the woman’s face, age, hair, cream blouse, earrings, pose, rooftop background and golden-hour lighting consistent with the starting image. Natural restrained facial motion.` ##### Kling 4.0 Lost the Portrait to Kling O3 Pro Kling 4.0 got the hard part right, also the thin strands of hair at the back looked realistic, and we did not see any artifacts or odd textures on the skin. The problem was the smile, which opened into a wide, open-mouth grin that made her look like she was on acid when the prompt asked for something subtle. Kling O3 Pro gave her a closed-mouth smile with the same prompt, and that read as far more natural. It is strange to watch the older model win, but on this portrait it did. 0:00 /0:05 1× Kling 4.0 handled the fine hair well, but its wide smile looked less natural than Kling O3 Pro’s. #### See the Kling 4.0 portrait prompt and settings Settings: 5 sec, 720p, 16:9, audio on, one starting image, prompt enhancer off Prompt: SUBJECT ACTION: The woman maintains direct eye contact, blinks naturally once and slowly forms a subtle warm smile. She turns her face slightly toward the sunlight. Only a few loose strands of hair move in the gentle breeze. CAMERA: Very slow cinematic dolly-in. One continuous shot. PRESERVATION: Keep the woman’s face, age, hair, cream blouse, earrings, pose, rooftop background and golden-hour lighting consistent with the starting image. Natural restrained facial motion. #### Veo 3.1 Looked Cinematic but Added Artificial Texture Veo surprised us with hair that moved more naturally than we expected, and its head turn gave the shot a slightly more cinematic feel than Kling’s. A closer look at the forehead showed fine artificial lines across the skin that did not read as pores or normal wrinkles, more like sharpening noise or a texture the model had invented while rebuilding the moving face. It was still a good clip, and in the portrait test Veo and Kling were much closer than the final ranking makes them sound. 0:00 /0:06 1× **Veo handled the hair well, but close inspection revealed an artificial texture across the face.* #### See the Veo portrait prompt and settings ****Settings:** 6 sec, 1080p, 16:9, audio off, default seed, prompt enhancer off ****Prompt:** `A continuous cinematic close-up at golden hour. The woman maintains eye contact with the lens and blinks naturally. Over the next few seconds, she gives a small, genuine smile and turns her face slightly toward the warm sunlight. A light breeze gently moves several strands of her hair. The camera slowly dollies forward while the rooftop background remains softly out of focus. Her appearance and the original composition remain consistent. Natural facial performance, understated fashion-film realism, no dialogue and no cuts.` #### Wan 2.7 Pro Could Not Join the Same Fight Wan 2.7 Pro was obviously weaker, with red marks across the skin, highlights that blew out and a face that looked more cartoonish than the source. The hair was passable, but the skin and eyes made the AI origin hard to ignore. Two years ago this would have looked impressive, but against the rest of this test we would not use it for a beauty advert or a close human portrait. Wan is popular for a reason, and we still see it in experimental ComfyUI workflows, including some completely unhinged character videos on Twitter, but that is a different job from a clean commercial portrait, where it finished last. 0:00 /0:05 1× **Wan introduced harsh facial texture and red marks that were not present in the source.* #### See the Wan portrait prompt and settings ****Settings:** 5 sec, 1080p, 16:9, audio on, default seed, prompt enhancer off ****Prompt:** `The woman looks directly into the camera. She blinks once naturally, gives a small warm smile, and turns her head slightly toward the sunlight. A gentle breeze moves a few strands of her hair. Slow camera push-in. One continuous realistic fashion portrait shot. Keep her facial appearance, hair, cream blouse, jewellery, pose, lighting and rooftop background consistent.` ## Test 2: Walking by the Pool Tested the Whole Body The second source gave the models a full body to handle, with open shoes, moving legs, a loose dress, sunglasses and a woven bag. Nano Banana Pro added the accessories without being asked, which made the test more useful, since every model now had two extra objects to keep intact while she walked. Her face is small in this frame and the eyes are not perfect before anything moves, and a video model cannot recover detail the image generator never created. The same untouched source went into all six models, though, so the comparison still holds. The poolside woman became Diana about ten seconds into our notes, so Diana she stayed. ![Blonde woman walking beside a luxury hotel swimming pool in bright Mediterranean daylight.](https://storage.ghost.io/c/21/2f/212fa2d6-dfa3-4908-9426-ec8fc852e19d/content/images/2026/08/TEST_SOURCE_2.jpg) **The 4K poolside source. The small face and added accessories made this a harder consistency test.* #### See the walking source-image prompt and settings ****Model:** Nano Banana Pro ****Settings:** 16:9, 4K, highest clean quality available ****Prompt:** `Photorealistic full-body fashion photograph of a beautiful 25-year-old blonde woman walking beside a luxury hotel swimming pool in bright Mediterranean daylight. She wears a fitted pale-blue summer dress and simple cream heels. Athletic hourglass figure, natural body proportions, loose shoulder-length hair and a relaxed confident expression. Captured mid-step from a slight front-side angle, with both hands clearly visible. Clean modern architecture, blue water and soft palm shadows in the background. Premium candid fashion campaign, realistic anatomy, horizontal 16:9 composition.` #### Seedance 2.5 Kept the Body, Clothes and Blink Together Seedance gave Diana a natural stride, stable shoes and convincing movement through the dress, and because she blinked near the end instead of staring through the camera for five seconds, the small expression change made her feel alive. The audio it generated was included in the $1.80 result, but it was wrong, with ambience closer to the sea than to a hotel pool, and a pool does not usually arrive with its own coastline. We left audio out of the ranking because the models were not run with matching sound settings, though when a model creates sound that gives the scene away, it is fair to mention it. 0:00 /0:05 1× **Seedance gave the most cohesive combination of gait, clothing, shoes and facial movement.* #### See the Seedance walking prompt and settings ****Settings:** 5 sec, 720p, 16:9, audio on, no last image, prompt enhancer off ****Prompt:** `5-second continuous tracking shot beside the pool. The woman naturally continues the walking stride already shown in the starting image, taking three relaxed steps forward. She keeps a soft confident smile and briefly glances toward the camera. Her handbag and the sunglasses in her hand swing subtly with her walking rhythm. Her pale-blue dress and hair respond gently to the breeze. The camera tracks smoothly beside her at the same slight front-side angle. Preserve her face, body proportions, dress, heels, handbag, sunglasses and the entire poolside setting. Realistic foot placement and fabric physics, with no cut or scene change.` #### Kling O3 Pro Nearly Won the Walking Test Kling O3 Pro’s walk looked genuinely good, with the open shoes staying on her feet, her weight shifting properly and background bokeh that made it feel like a real fashion clip rather than a moving still. The one oddity was a stare that lasted almost five seconds without a blink. Maybe she was flirting with us, but once we noticed it, it felt slightly uncanny. For motion and value this was one of Kling’s best results, and we would happily use it after trimming the end or hiding the stare inside a normal edit. 0:00 /0:05 1× **Kling handled the gait and open shoes extremely well, although Diana forgot to blink.* #### See the Kling walking prompt and settings ****Settings:** 5 sec, 1080p, 16:9, audio off, one starting image, prompt enhancer off ****Prompt:** SUBJECT ACTION: Continue the woman’s existing walking motion. She takes three natural relaxed steps beside the pool and briefly glances toward the camera while keeping a soft smile. Her handbag and sunglasses swing slightly with each step. Her dress and hair move gently in the breeze. CAMERA: Smooth lateral tracking shot, maintaining the same front-side viewing angle and full-body framing. One continuous take. `PRESERVATION: Keep her face, body proportions, pale-blue dress, cream heels, handbag, sunglasses and poolside environment consistent with the starting image. Realistic walking mechanics and grounded feet.` ##### Kling 4.0 Matched Kling O3 Pro on the Walk Kling 4.0 gave Diana a natural walk where her face did not morph, the hands and glasses stayed in place and her hair moved normally in the wind, while the camera followed her smoothly like a real tracking shot. We watched it back to back with Kling O3 Pro and could not find a meaningful difference in the movement. Both are very strong here, and Seedance still had the more complete performance. 0:00 /0:05 1× Kling 4.0 kept Diana’s face, glasses and gait stable, with a smooth following camera. #### See the Kling 4.0 walking prompt and settings Settings: 5 sec, 720p, 16:9, audio on, one starting image, prompt enhancer off Prompt: SUBJECT ACTION: Continue the woman’s existing walking motion. She takes three natural relaxed steps beside the pool and briefly glances toward the camera while keeping a soft smile. Her handbag and sunglasses swing slightly with each step. Her dress and hair move gently in the breeze. CAMERA: Smooth lateral tracking shot, maintaining the same front-side viewing angle and full-body framing. One continuous take. PRESERVATION: Keep her face, body proportions, pale-blue dress, cream heels, handbag, sunglasses and poolside environment consistent with the starting image. Realistic walking mechanics and grounded feet. ##### Wan 3.0 Walked Cleanly, With a Slight Cartoon Edge Along with music that sounded a lot like Veo’s, Wan 3.0 added a soft glow over the scene, and although Diana pulled a few funny expressions, her hair, shoes and body held together without any morphing. The weakness only showed on her legs, where faint outlines give the clip a slightly cartoon look. You have to nitpick to see it, but next to Seedance it is there, and Seedance stayed ahead in this test. 0:00 /0:05 1× Wan 3.0 kept the walk and accessories intact, with faint cartoon outlines around the legs. #### See the Wan 3.0 walking prompt and settings Settings: 5 sec, 720p, 16:9, audio on, prompt expansion off, no last image Prompt: The woman continues walking naturally from the pose in the starting image, taking three relaxed forward steps beside the pool. She briefly looks toward the camera with a soft smile. Her handbag and sunglasses swing gently with her movement. Her dress and hair move slightly in the breeze. Smooth side-tracking camera at the same front-side angle. Keep her face, body, clothing, accessories and poolside background consistent. Realistic walking and foot contact with the ground. #### Veo 3.1 Looked Like It Had Cut Her Out of the Scene Veo got the broad action right: the bag survived, the walk made sense and most of the glasses stayed intact until they started changing near the end. What bothered us was the edge around her moving body, which looked as if the model had cut Diana out, tracked that shape and slid it through the background. We do not know whether Veo works anything like that inside, but that is what the visible contour made the finished video look like. With the hair also behaving strangely and the glasses losing their shape late in the clip, this was where Veo clearly dropped behind Kling. 0:00 /0:06 1× **Veo completed the walk, but the moving body gained a visible cutout-like edge and the glasses weakened near the end.* #### See the Veo walking prompt and settings ****Settings:** 6 sec, 1080p, 16:9, audio off, default seed, prompt enhancer off ****Prompt:** `A continuous full-body fashion tracking shot beside a Mediterranean hotel pool. Beginning from her existing mid-step position, the woman walks forward naturally for three relaxed steps. She keeps a soft smile and briefly looks toward the camera. The handbag and sunglasses already in her hands move subtly with her stride, while her dress and hair react gently to the breeze. The camera tracks alongside her at the same slight front-side angle, maintaining stable full-body framing. Realistic gait, grounded footfalls, natural fabric movement and a consistent poolside background. No dialogue and no cuts.` #### Wan 2.7 Pro Sent Diana on an Acid Trip Wan 2.7 Pro completed the basic movement, but the face, nose and eyes drifted as she walked, and the shoes and body were less consistent too. It reminded us of an older generation of AI video where everything looks acceptable for half a second and then begins softly changing into something else. We cannot blame the source for this one, because every other model got the same small face and kept it together far more convincingly. 0:00 /0:05 1× **Wan understood “walk forward,” but struggled to preserve the same person through the movement.* #### See the Wan walking prompt and settings ****Settings:** 5 sec, 1080p, 16:9, audio on, default seed, prompt enhancer off ****Prompt:** `The woman continues walking naturally from the pose in the starting image, taking three relaxed forward steps beside the pool. She briefly looks toward the camera with a soft smile. Her handbag and sunglasses swing gently with her movement. Her dress and hair move slightly in the breeze. Smooth side-tracking camera at the same front-side angle. Keep her face, body, clothing, accessories and poolside background consistent. Realistic walking and foot contact with the ground.` ## Test 3: The Perfume Advert Separated the Best Two The final source looked closest to a real commercial frame. One hand hovered above a clear perfume bottle while the other rested on the table. We asked each model to pick up the bottle, raise it to chest level, follow it with her eyes and give a small satisfied smile. This gave the models several opportunities to embarrass themselves. Fingers had to close around glass. The bottle needed to remain rigid and transparent. Reflections and liquid had to move with it. Her gaze needed to follow the object without the face falling apart. ![East Asian woman seated at a minimalist cream dressing table](https://storage.ghost.io/c/21/2f/212fa2d6-dfa3-4908-9426-ec8fc852e19d/content/images/2026/08/TEST_SOURCE_3.jpg) **The untouched 4K perfume source used for all six product-interaction videos.* #### See the perfume source-image prompt and settings ****Model:** Nano Banana Pro ****Settings:** 16:9, 4K, highest clean quality available ****Prompt:** `Photorealistic luxury beauty advertisement featuring a beautiful 27-year-old East Asian woman seated at a minimalist cream dressing table. She wears an elegant black sleeveless evening dress and refined gold earrings. One hand rests beside a rectangular clear-glass perfume bottle with a gold cap, while the other rests naturally on the table. Both hands and all fingers are clearly visible. She looks toward the perfume with a calm, sophisticated expression. Soft warm studio lighting, realistic skin texture, clean cream background, premium cosmetics campaign, horizontal 16:9 composition.` #### Seedance 2.5 Made the Best Advert This was Seedance at its best as always. We liked how her fingers closed around the bottle naturally, the perfume remained convincing and the eyes followed the movement. Then came a tiny smile that felt like a human reaction rather than an expression slider being moved from neutral to happy. Light travelled through the glass and onto her hand. The liquid remained believable. The bottle did not melt, duplicate or decide it wanted a new shape halfway up. Just keep feeding Seedance money and, apparently, it keeps doing things like this. 0:00 /0:05 1× **Seedance combined stable hands, believable glass and the best micro-expression of the test.* #### See the Seedance perfume prompt and settings ****Settings:** 5 sec, 720p, 16:9, audio on, no last image, prompt enhancer off ****Prompt:** `5-second continuous luxury beauty-commercial shot. The woman lowers her raised hand toward the perfume bottle, closes her fingers naturally around the bottle and lifts it smoothly from the dressing table to approximately chest level. She looks at the bottle and gives a restrained, satisfied smile. The rectangular clear-glass bottle and gold cap remain rigid, transparent and unchanged throughout the action. Her other hand remains relaxed on the table. The camera makes a very slow controlled move from left to right. Preserve her face, black dress, jewellery, hands, bottle design, table and cream studio environment. Elegant realistic movement, no cut or scene change.` ##### Wan 3.0 Made the Most Interesting Perfume Clip Wan 3.0 surprised us again by moving the camera round to show her from the front as she picked up the bottle and lifted it in one natural motion, with the liquid moving inside and her eyes going from straight ahead to the perfume at a new angle. This was the most interesting of the six perfume clips to watch, although Seedance kept the edge on the face and the small smile, so it stays first. With the leading models this close, you have to look carefully to separate them. 0:00 /0:05 1× Wan 3.0 lifted the bottle naturally and gave the most interesting camera move of the test. #### See the Wan 3.0 perfume prompt and settings Settings: 5 sec, 720p, 16:9, audio on, prompt expansion off, no last image Prompt: The woman lowers her raised hand toward the perfume bottle, grips it naturally and lifts it smoothly from the table to chest level. She looks at the bottle and gives a subtle smile. Her other hand stays resting on the table. The rectangular transparent glass bottle and gold cap remain rigid and unchanged. Slow controlled camera movement from left to right. Keep her face, hands, black dress, jewellery, table, bottle and cream studio background consistent. Realistic hand and object interaction. #### Kling O3 Pro Was Ridiculously Good for the Price Kling O3 Pro followed the instruction closely, picking up the bottle, watching it and keeping the other hand in place without losing fingers or turning the perfume into soup. What impressed us most were the small details, like the eyes moving toward the product, the shadows changing properly and the light through the bottle leaving a believable reflection on her hand. The reaction was a little stiff, but that may partly be our fault, because we wrote a highly controlled AI prompt and received a highly controlled AI perfume model. This is the strongest argument for Kling, since it was behind Seedance but nowhere near three times worse. 0:00 /0:05 1× **Kling obeyed the action, preserved the bottle and handled the hand contact far better than its price suggests.* #### See the Kling perfume prompt and settings ****Settings:** 5 sec, 1080p, 16:9, audio off, one starting image, prompt enhancer off ****Prompt:** SUBJECT ACTION: The woman lowers her raised hand, wraps her fingers naturally around the rectangular perfume bottle and lifts it smoothly from the table to chest level. She follows the bottle with her eyes and gives a small satisfied smile. Her other hand remains resting on the table. OBJECT BEHAVIOUR: The clear-glass perfume bottle and gold cap remain rigid, transparent and identical in size and shape. The bottle does not bend, melt or change design. CAMERA: Very slow lateral camera move from left to right. One continuous luxury beauty-commercial shot. `PRESERVATION: Keep her face, black dress, jewellery, hands, dressing table and cream studio background consistent with the starting image.` ##### Kling 4.0 Got the Mirror and the Bottle Right Ignoring the sound, Kling 4.0 did a very good job, with a mirror reflection that held up better than in any other perfume clip, a bottle that kept its size and something visibly moving inside the glass as she lifted it. Next to Kling O3 Pro the pose looked more natural, but the older model’s image quality was better. That is an odd result for a newer model, and it is why Kling 4.0 finished behind its older brother overall. 0:00 /0:05 1× Kling 4.0 kept the bottle’s size and gave the best mirror reflection, but Kling O3 Pro looked sharper. #### See the Kling 4.0 perfume prompt and settings Settings: 5 sec, 720p, 16:9, audio on, one starting image, prompt enhancer off Prompt: SUBJECT ACTION: The woman lowers her raised hand, wraps her fingers naturally around the rectangular perfume bottle and lifts it smoothly from the table to chest level. She follows the bottle with her eyes and gives a small satisfied smile. Her other hand remains resting on the table. OBJECT BEHAVIOUR: The clear-glass perfume bottle and gold cap remain rigid, transparent and identical in size and shape. The bottle does not bend, melt or change design. CAMERA: Very slow lateral camera move from left to right. One continuous luxury beauty-commercial shot. PRESERVATION: Keep her face, black dress, jewellery, hands, dressing table and cream studio background consistent with the starting image. #### Veo 3.1 Understood That It Was an Advert Veo made a decent and neat commercial, not gonna lie, because after lifting the bottle and looking at it she turned and presented it to the camera with a bigger smile, as if she understood the purpose of the shot and decided to direct the ending herself. That made it attractive and less faithful at the same time, since we asked for the bottle at chest level and a restrained reaction, and Veo raised it higher and turned the moment into a product presentation. The eyes and teeth held together, so the weak part was the same artificial edge and glow from the walking test, especially around the subject and skin, and with the clip already at 1080p, resolution cannot explain it away. 0:00 /0:06 1× **Veo produced a convincing ad concept, although it directed a more presentational ending than we requested.* #### See the Veo perfume prompt and settings ****Settings:** 6 sec, 1080p, 16:9, audio off, default seed, prompt enhancer off ****Prompt:** `A continuous luxury beauty-commercial shot in a warm cream studio. The woman lowers her raised hand toward the rectangular perfume bottle, closes her fingers naturally around it and lifts it from the dressing table to chest level. Her eyes follow the bottle and she gives a subtle, satisfied smile. Her other hand remains resting naturally on the table. The clear-glass bottle and gold cap retain exactly the same rigid shape, scale and transparent material throughout the movement. The camera makes a slow, precise lateral move from left to right. Refined restrained performance, realistic hand-object contact, no dialogue and no cuts.` #### Wan 2.7 Pro Looked Like She Wanted to Eat the Bottle Wan 2.7 Pro also picked up the perfume and generated sound, but the quality fell as soon as the action developed. The eyes weakened, the bottle changed scale, the liquid became an unclear glowing movement inside the glass, and her second hand joined the bottle even though we had asked it to stay on the table. After an opening that was not terrible, the hair began moving as a few stiff, drawn shapes, the finger folds turned strange and the whole interaction started to look like she was wondering whether the perfume might be lunch. Wan 2.7 Pro was cheaper than Seedance and Veo, but even its starting price was above Kling O3 Pro’s $0.56, and Wan 3.0 now does better for less. 0:00 /0:05 1× **Wan completed the idea, but the second hand joined in and the bottle lost its original material and scale.* #### See the Wan perfume prompt and settings ****Settings:** 5 sec, 1080p, 16:9, audio on, default seed, prompt enhancer off ****Prompt:** `The woman lowers her raised hand toward the perfume bottle, grips it naturally and lifts it smoothly from the table to chest level. She looks at the bottle and gives a subtle smile. Her other hand stays resting on the table. The rectangular transparent glass bottle and gold cap remain rigid and unchanged. Slow controlled camera movement from left to right. Keep her face, hands, black dress, jewellery, table, bottle and cream studio background consistent. Realistic hand and object interaction.` ## Seedance Made the Best Videos, but Kling Is the Smarter Buy Our final order is Seedance 2.5, then Wan 3.0, Kling O3 Pro, Kling 4.0, Veo 3.1 and Wan 2.7 Pro. Seedance is the quality winner, with the most *realest* and alive portrait eyes, the most complete walk from Diana and the best perfume interaction. ByteDance only [released Seedance 2.5](https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5?ref=theseguysknow.io) on July 31, 2026, and it already feels like a real step ahead for short human clips. For ordinary social clips or trying several ideas with our own money, we would now open Wan 3.0 first, because three Wan 3.0 generations cost $1.50 while three Seedance generations cost $5.40\. Wan 3.0 lost every direct comparison with Seedance, but it beat both Kling models on the portrait and made the most interesting perfume clip. The two Klings were the hardest pair to separate, with almost identical walking clips, a more natural perfume pose from Kling 4.0 and a better portrait and cleaner image from Kling O3 Pro. We put the older model ahead because it won the close-up and costs less, at $0.56 for 1080p against $0.79 for 720p. Veo remains interesting, but it sometimes made choices that felt more like direction than literal prompt following, and the artificial skin texture and the edge around the moving body stopped us recommending it above either Kling, especially because these tests were already rendered at 1080p. Google officially supports four, six and eight-second outputs at both 720p and 1080p in its [Veo 3.1 specifications](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/veo/3-1-generate?ref=theseguysknow.io). Wan 2.7 Pro is harder to recommend for this job, and Wan 3.0 is the obvious upgrade. Alibaba’s [Wan 2.7 model](https://www.alibabacloud.com/help/en/model-studio/image-to-video-general-api-reference?ref=theseguysknow.io) can handle first-frame generation, first-and-last-frame video and continuation, so this small test does not cover its full range, but for faces, walking and a clean beauty advert it was visibly behind and did not even undercut Kling O3 Pro. Seedance is still worth the money when we want the best first result we can reasonably buy, or when we have a strange idea and want to see how far the model will take it. There are so many things left to try that we are almost afraid to think what these videos will look like next year. For the work we produce every week, Wan 3.0 is easier to justify, and it got close enough that we would rather make three Wan 3.0 attempts than one Seedance attempt in many situations. When the face, emotion or hero product shot matters most, we would still pay extra for Seedance. ## FAQ #### What is the best AI image-to-video generator in 2026? Seedance 2.5 made the best clips in our tests, especially for faces and product shots, while Wan 3.0 is the best value at $0.50 per five-second 720p clip on WaveSpeed. #### Is Kling 4.0 better than Kling O3 Pro? Not in our test, because Kling O3 Pro won the portrait, matched Kling 4.0 on the walk and costs less per clip, even though Kling 4.0 had realistic hair and a more natural perfume pose. #### Which image-to-video model is the cheapest? Of the six we tested, Wan 3.0 is the cheapest at $0.50 per five-second 720p clip before discounts, and it finished second overall. #### Does the source image matter for image-to-video? It matters a lot, because the video model builds on the character, lighting and detail in the image, so a small or soft face in the source stays soft in the video. Every model here started from the same 4K Nano Banana Pro images. #### External Sources - [Google DeepMind: Nano Banana Pro](https://deepmind.google/models/gemini-image/pro/?ref=theseguysknow.io) - [ByteDance Seed: Introducing Seedance 2.5](https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5?ref=theseguysknow.io) - [Kling AI: VIDEO 3.0 Omni model guide](https://app.klingai.com/global/quickstart/klingai-video-3-omni-model-user-guide?ref=theseguysknow.io) - [Google: Veo 3.1 specifications](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/veo/3-1-generate?ref=theseguysknow.io) - [Alibaba Cloud: Wan 2.7 image-to-video API](https://www.alibabacloud.com/help/en/model-studio/image-to-video-general-api-reference?ref=theseguysknow.io) - [WaveSpeed: Seedance 2.5 endpoint and pricing](https://wavespeed.ai/models/bytedance/seedance-2.5/image-to-video?ref=theseguysknow.io) - [WaveSpeed: Kling O3 Pro endpoint and pricing](https://wavespeed.ai/models/kwaivgi/kling-video-o3-pro/image-to-video?ref=theseguysknow.io) - [WaveSpeed: Veo 3.1 endpoint and pricing](https://wavespeed.ai/models/google/veo3.1/image-to-video?ref=theseguysknow.io) - [WaveSpeed: Wan 2.7 Pro endpoint and pricing](https://wavespeed.ai/models/alibaba/wan-2.7/image-to-video-pro?ref=theseguysknow.io)