Real videos generated by the OpenX Flow pipeline. Every clip below was created autonomously from a text prompt — no human editing, no cherry-picking.
Cost per scene
Cost per minute
Generation time
Total cost (9 clips)
A 2.5-minute narrated short film about a forgotten library at the edge of the world where ancient books come alive at night. 30 scenes generated autonomously — script, narration, video, and assembly. Total cost: $1.46.
30
2:27
$1.46
Wan2.2-TI2V-5B
A100-40GB
50
832×480
Aerial: rust-orange dunes, lone library
Push-in: ancient stone library
Close-up: weathered stone carvings
Interior: vast towering bookshelves
Hundreds of books glowing
Stars burst from open book
Luminous tree grows from book
Tree dissolves into golden dust
Final: library under stars
Full pipeline demo — AI narration over AI-generated deep sea footage. Script → TTS → Video → Assembly. The narration was generated for free with Edge TTS, then overlaid on the video with ffmpeg.
3
15s
$0.25
Wan2.2-TI2V-5B
A100-40GB
50
832×480
Descending into dark ocean, fading sunlight
Bioluminescent jellyfish pulsing in darkness
Anglerfish with glowing lure in the abyss
A lone astronaut explores the Martian surface — walking, discovering a rock, storing it, and departing at sunset. 5 connected scenes generated autonomously from text prompts.
5
25s
$0.30
Wan2.2-TI2V-5B
A100-40GB
50
832×480
Astronaut walks across red desert
Looks up at Earth in the sky
Picks up iridescent rock
Stores specimen, camera pulls back
Wide shot — walks toward lander at sunset
A water droplet falls, hits a surface creating ripples, then the camera pulls back to reveal an ocean at golden hour. Cinematic macro to wide transition.
3
15s
$0.26
Wan2.2-TI2V-5B
A100-40GB
50
832×480
Macro: single droplet falling, black bg
Droplet impacts water, concentric ripples
Wide: vast ocean at golden hour sunset
| Model | Wan2.2-TI2V-5B (5B params) |
| GPU | NVIDIA A100-40GB (Modal) |
| Resolution | 832×480 (16:9) |
| Frames per clip | 81 (5.06s @ 16fps) |
| Inference steps | 50 |
| Guidance scale | 5.0 |
| Precision | FP16 |
| 30 steps | 51.3s/clip — faster, slightly lower quality |
| 50 steps | 82.8s/clip — best quality |
| Speedup | 1.6× for 30 steps |
30 steps is viable for drafts. 50 steps for final output.
20 diverse scenes generated in a single batch. Total cost: $1.03. Themes: underwater, space, nature, abstract, urban.
All benchmarks are reproducible. Clone the repo and run with your own Modal account.
View Benchmark Code on GitHub