One of the most challenging aspects in AI-generated videos is realism — many things need to align for the video to be realistic, from motion to the level of details, from colors to lighting and the overall world building. AI video generation models have made a lot of progress in recent years, but can they create truly realistic videos and if yes — which models can? To answer this question, we tested 11 AI video models on the same challenging prompt. Here's what we found.
We used the following prompt to compare the models:
Format & Look
Duration 4s; 180° shutter; digital capture emulating 65 mm photochemical contrast; fine grain; subtle halation on speculars; no gate weave.
Lenses & Filtration
32 mm / 50 mm spherical primes; Black Pro-Mist 1/4; slight CPL rotation to manage glass reflections on train windows.
Grade / Palette
Highlights: clean morning sunlight with amber lift.
Mids: balanced neutrals with slight teal cast in shadows.
Blacks: soft, neutral with mild lift for haze retention.
Lighting & Atmosphere
Natural sunlight from camera left, low angle (07:30 AM).
Bounce: 4×4 ultrabounce silver from trackside.
Negative fill from opposite wall.
Practical: sodium platform lights on dim fade.
Atmos: gentle mist; train exhaust drift through light beam.
Location & Framing
Urban commuter platform, dawn.
Foreground: yellow safety line, coffee cup on bench.
Midground: waiting passengers silhouetted in haze.
Background: arriving train braking to a stop.
Avoid signage or corporate branding.
Wardrobe / Props / Extras
Main subject: mid-30s traveler, navy coat, backpack slung on one shoulder, holding phone loosely at side.
Extras: commuters in muted tones; one cyclist pushing bike.
Props: paper coffee cup, rolling luggage, LED departure board (generic destinations).
Sound
Diegetic only: faint rail screech, train brakes hiss, distant announcement muffled (-20 LUFS), low ambient hum.
Footsteps and paper rustle; no score or added foley.
Optimized Shot List (2 shots / 4 s total)
0.00–2.40 — "Arrival Drift" (32 mm, shoulder-mounted slow dolly left)
Camera slides past platform signage edge; shallow focus reveals traveler mid-frame looking down tracks. Morning light blooms across lens; train headlights flare softly through mist. Purpose: establish setting and tone, hint anticipation.
2.40–4.00 — "Turn and Pause" (50 mm, slow arc in)
Cut to tighter over-shoulder arc as train halts; traveler turns slightly toward camera, catching sunlight rim across cheek and phone screen reflection. Eyes flick up toward something unseen. Purpose: create human focal moment with minimal motion.
Camera Notes (Why It Reads)
Keep eyeline low and close to lens axis for intimacy.
Allow micro flares from train glass as aesthetic texture.
Preserve subtle handheld imperfection for realism.
Do not break silhouette clarity with overexposed flare; retain skin highlight roll-off.
Finishing
Fine-grain overlay with mild chroma noise for realism; restrained halation on practicals; warm-cool LUT for morning split tone.
Mix: prioritize train and ambient detail over footstep transients.
Poster frame: traveler mid-turn, golden rim light, arriving train soft-focus in background haze.
The prompt is the ultra-detailed example from OpenAI's Sora 2 prompting guide. All clips were generated in 16:9, at 1080p where the model supports it. The prompt was identical for all models, except for Kling 3 and Hailuo 2.3 Pro, which have a cap of 2,500 and 2,000 characters respectively. These two models received a shortened prompt.
We compared the following models, listed from best to worst in the table below:
| Rank | Model | Developer | Result | Resolution |
|---|---|---|---|---|
| 1 | Kling 3 Turbo Pro | Kuaishou | Hard to tell from real footage | 1080p |
| 2 | Flux 3 | Black Forest Labs | Realistic crowd and camera movement | 1080p |
| 3 | Kling 3 | Kuaishou | Realistic, with artifacts in the crowd | 720p |
| 4 | Seedance 2.0 | ByteDance | Close to Kling 3 Turbo Pro, with less detail | 1080p |
| 5 | Seedance 2.5 | ByteDance | Realistic shots, lighting changes between them | 1080p |
| 6 | Grok Imagine Video | xAI | Realistic motion, smoothed look | 720p |
| 7 | Veo 3.1 | Plastic textures, unnatural motion | 1080p | |
| 8 | MiniMax H3 | MiniMax | Oversaturated, waxy details | 2K |
| 9 | Hailuo 2.3 Pro | MiniMax | Oversaturated, objects in illogical places | 1080p |
| 10 | Wan 2.6 | Alibaba | Plastic look | 1080p |
| 11 | Hunyuan Video 1.5 | Tencent | Overexposed, simple composition | 480p |
By the way, all of these models are available in the Overchat AI video generator, our all-in-one AI app, so if you want to test them yourself, you can do so there.
Kling 3 Turbo Pro by Kuaishou produced the most realistic clip in the test.
The clip has a high level of detail and natural motion, and the image stays stable from start to finish. On a quick viewing, nothing in it indicates AI generation. Kling 3 Turbo Pro also followed the prompt most closely. The clip includes the cyclist, the coffee cup on the bench, the yellow safety line and the phone in the traveler's hand, and the cut to the close-up happens at 2.96 seconds, close to the 2.40 seconds in the shot list.
The clip has a dense crowd, detailed surroundings and atmospheric perspective created by the haze.
The model isn't as accurate at instruction following, perhaps — the cut to the close-up happens at about 4 seconds, later than the shot list specifies.
Kling 3 by Kuaishou produced a clip close in realism to Kling 3 Turbo Pro, if not for some obvious AI artifacts.
Watch the man moving away from the camera — halfway through he dissolves into nothingness and morphs into a man moving towards the camera. Also, the clip is a single shot without the cut from the shot list, and this model maxes out resolution at 720p.
Seedance 2.0 by ByteDance produced a clip similar to the one from Kling 3 Turbo Pro.
The clip has less detail and less natural motion than the Kling 3 Turbo Pro clip. It also contains a sign with unreadable text, although the prompt says to avoid signage.
Seedance 2.5 by ByteDance produced one of the most realistic images in the test, with dense fog and a close-up of the traveler's face.
After the cut at 2.71 seconds, the lighting changes, so the two shots don't match. For this reason, Seedance 2.5 is better suited for generating one shot at a time.
Grok Imagine Video by xAI produced realistic motion of the traveler and the passengers.
The image looks smoother and has fewer details than the Flux 3 clip, so the overall result is less realistic. The departure board on the train shows unreadable text. The maximum resolution of the model is 720p.
Veo 3.1 by Google produced a clip with warm light and lens flares.
The textures look plastic and the movements look unnatural. The clip is a single 4-second shot without a cut, and large green flares cover part of the frame.
MiniMax H3 produced a clip with good camera work.
However, the colors are oversaturated, the platform is nearly empty, and blurred, waxy details, especially in the skin, indicate AI generation.
Hailuo 2.3 Pro by MiniMax produced the most stylized clip in the test.
The colors are oversaturated, the crowd is sparse and contains artifacts, and some objects are placed illogically. The yellow safety line appears as yellow tape stretched across the platform, and the train moves to the other side of the frame within one continuous shot.
Wan 2.6 by Alibaba produced a clip with a plastic look, which is not suitable for realistic video.
The cut to the second shot happens at 2.30 seconds, the closest to the shot list in the test. In the second shot, the traveler drinks coffee while looking at the camera, an action that the prompt doesn't describe, and the platform has maps and boards with text.
Hunyuan Video 1.5 by Tencent produced the least realistic clip in the test.
The image is overexposed, the composition is the simplest of all the clips, and the dawn atmosphere is missing. The traveler holds the phone to his ear, although the prompt describes the phone held at his side. The clip is 480p and has no sound.
In our test, Kling 3 Turbo Pro produced the most realistic video, with a high level of detail, a stable image and natural motion. Flux 3 and Kling 3 took second and third place.
In our test, the most common signs of AI generation were plastic or waxy textures (Wan 2.6, Veo 3.1, MiniMax H3), oversaturated colors (MiniMax H3, Hailuo 2.3 Pro), unreadable text on signs (Seedance 2.0, Grok Imagine Video, Wan 2.6), lighting that changes between shots (Seedance 2.5) and distorted people in the background (Kling 3, Hailuo 2.3 Pro).
Kling 3 Turbo Pro followed the test prompt most accurately: the clip includes every requested element, and the cut between the two shots happens close to the requested time. Wan 2.6 matched the cut timing most precisely but added an action that the prompt doesn't describe.
In Overchat AI, video models generate clips of 5 or 10 seconds. Longer videos are assembled from several clips or made longer with an AI video extender.
Yes, with limits. Several free AI video generators let you create a few clips without paying.
