This is an AI video generation comparison for image-to-video 
prompt:
Surreal, imaginative yet realistic scene shows a man climbing through a magical portal shaped like a vintage TV into a beach out of his poor looking boring appartment. His arm is seen crossing the boundary and touching the sand inside a TV frame before he climbs further inside crossing into the beach side
Tested: July 2, 2025
Pika has been resisting the physics-bending part, but this one generation comes close
Tested: July 2, 2025
Kling didn't muck around and sent the guy into the beach side straight away
Tested: July 3, 2025
One out of 4 results is decent, most have a bit of awkwardness to the man's movements.
Tested: July 19, 2025
Well, it followed the prompt, there was just a bit of inconsistent leg motion going on that isn't even easy to spot at first glance.
Is the man’s pose anatomically correct with no visible warping or limb issues as he climbs through?
Does the surreal transition (TV as portal) look physically plausible and integrated (no floating elements or jagged seams)?
Is there a clear sense of depth showing both interior space (apartment) and exterior (beach) coexisting through the portal?
Is the scene grounded in realistic detail despite the surreal concept (textures, lighting, material physics)?
P.S. The image idea isn't mine, was kindly shared on Reddit a while ago, I've regenerated it to test out this surreal prompt.
Check out the results from Dreamina AI (Seedance 1.0 Pro) vs Pika (Pika 2.2) vs Hailuo AI x Minimax (Hailuo 02) vs PixVerse (Pixverse V4.5) vs Wan (Online Platform) (Wan 2.1) vs Kling AI (Kling 2.1) vs Midjourney (Midjourney V1) vs LTX-Video by Lightricks (LTX Video 0.9.8-13B Distilled) for similar or identical prompts side-by-side.
