P-Video: text-to-video and image-to-video in 6 tests
P-Video generates short video clips from a text prompt, and it can also animate from a single input image. This post runs six prompts that look like real production requests: product ads, city b-roll, food motion, and action sports.
Model link
Test setup
- Resolution: 720p
- FPS: 24
- Duration: 5 seconds (one test at 7 seconds)
- Three runs were text-to-video. Two runs used a single input image. One run disabled prompt upsampling.
Run time snapshot
| Test | Mode | Ratio | Elapsed seconds |
|---|---|---|---|
| 01 | Text-to-video | 16:9 | 18 |
| 02 | Text-to-video | 16:9 | 16 |
| 03 | Text-to-video | 9:16 | 16 |
| 04 | Image-to-video | Input image | 25 |
| 05 | Image-to-video | Input image | 18 |
| 06 | Text-to-video | 16:9 | 21 |
Results
Test 01: product ad watch shot
Prompt:
Ultra realistic product ad. A stainless steel wristwatch on matte black marble. Slow rotating turntable. Macro lens. Crisp reflections. Subtle dust motes. Camera does a gentle push in. High end studio lighting.
Quick take: reflective materials look good when the prompt stays simple. Watch hands and bezel edges still need a close look frame by frame.
Test 02: rainy neon street b-roll
Prompt:
Cinematic neon street at night in heavy rain. A dark sedan drives past. Wet asphalt reflections. Neon signs blur in the background. Smooth tracking shot from left to right. Realistic raindrops on the lens.
Quick take: motion blur and reflections sell the vibe. Sign shapes and car geometry can wobble under heavy rain effects.
Test 03: vertical food motion
Prompt:
Top down food video. A steaming ramen bowl. Chopsticks lift noodles slowly. Steam swirls. Warm soft lighting. Shallow depth of field. Natural motion. Realistic textures.
Quick take: steam and liquid motion usually look best at shorter durations. Fingers and chopsticks are a common failure point in many video models.
Test 04: image-to-video action sports
Prompt:
Action sports. A snowboard instructor carves down a bright slope. Smooth dynamic tracking shot behind the rider. Crisp daylight. Snow spray in slow motion.
Quick take: image-to-video helps lock the starting composition. Fast motion still risks warping around boards and boots.
Test 05: image-to-video city street pan
Prompt:
Cinematic urban street at dusk. A sleek dark sedan moves across frame. Background traffic approaches camera. Warm street lamps. Slow steady pan tracking the sedan. Photorealistic.
Quick take: pans often stay stable if the prompt avoids extra scene changes. Watch for wheel rotation and headlight flicker.
Test 06: hard prompt, fantasy wildlife
Prompt:
High fantasy documentary. A majestic stag with mossy antlers in a foggy redwood forest. Bioluminescent spores floating. Slow camera push in from wide to medium close up. Golden hour god rays. Shallow depth of field.
Quick take: this prompt stacks multiple effects. The most common issue is temporal consistency on fine details like antlers and drifting particles.
Takeaways
- Short prompts with clear camera motion cues work best.
- Image-to-video can stabilize the first frame and composition.
- High detail + fog + particles pushes temporal consistency hardest.
