28/08/2026
I tested Kling 3.0, CUSTOM, WAN 2.7, and Veo 3.1 Lite to see how differently each system interprets the same character and scene.
And there’s something different about one of these tests:
CUSTOM is my own AI video creation app.
I’ve been building it to make AI video creation and experimentation more controlled and practical, especially for character consistency, reference images, motion direction, and repeatable workflows. It’s still evolving, but I’ll be opening CUSTOM for public testing soon.
And more importantly: Please tell me which AI performed best this time based on the prompt below?
PROMPT:
Use the attached images as the direct character and scene references.
Image roles:
Image 1 = starting framing and wider camera view
Image 2 = ending framing and closer zoomed-in view
Animate the same man seated in the same office. Preserve his identity, hairstyle, beard/stubble, teal shirt, pose, character scale, background, desk objects, backpack, shopping bag, lighting, and visual style.
Create a smooth realistic video where the camera begins with the framing of Image 1 and then slowly and smoothly zooms in until it reaches the closer framing of Image 2. The zoom must feel natural, steady, and cinematic, with no sudden change in scale or framing.
The computer monitor and keyboard are off-screen in front of him and must remain invisible. He is naturally working at the unseen computer. Animate small realistic finger, hand, wrist, and forearm movements suggesting typing, with occasional brief pauses. Add subtle shoulder movement, breathing, blinking, and small eye movements while his gaze remains focused on the off-screen monitor.
At the same time, he quietly enjoys music. Keep the reaction extremely subtle: tiny rhythmic head movement, minimal shoulder/upper-body response, and slight softening of the eyes. His expression should remain calm, focused, and inwardly absorbed. Do not add any smile. His work remains the primary action.
Keep his head, torso, seated position, and overall pose stable. No singing or lip-syncing; mouth mostly still.
Negative prompt:
no smile, no closed-mouth smile, no grin, no teeth showing, no exaggerated expression, no visible monitor or keyboard, no headphones, no dancing, no strong head bobbing, no large gestures, no looking at camera, no head turning, no standing, no body repositioning, no sudden zoom, no shaky camera, no background/object changes, no identity drift, no face distortion, no clothing change, no extra fingers, no warped hands, no flicker, no sudden motion.
