vllama run_video — Video Generation

Generate short videos from text prompts using text-to-video models.

---

Syntax

vllama run_video <model_name> [--prompt <text>] [--service <service>] [--output_dir <dir>]

---

Parameters

Parameter	Short	Default	Description
model_name		required	HuggingFace model ID
--prompt	-p	interactive	Text prompt for generation
--service	-s	local	Cloud service (kaggle)
--output_dir	-o	current dir	Where to save the output video

---

Examples

# Generate video locally
vllama run_video damo-vilab/text-to-video-ms-1.7b --prompt "A cat playing piano"

# Interactive mode
vllama run_video damo-vilab/text-to-video-ms-1.7b
# Prompt> A sunset over the ocean
# Prompt> exit

# Offload to Kaggle GPU (recommended — video gen is VRAM-heavy)
vllama run_video damo-vilab/text-to-video-ms-1.7b --service kaggle --prompt "A robot dancing"

# Custom output folder
vllama run_video damo-vilab/text-to-video-ms-1.7b -p "Waves crashing on shore" -o ./videos

---

Recommended Model

Model	Description
damo-vilab/text-to-video-ms-1.7b	ModelScope's 1.7B text-to-video model — solid quality for short clips

---

Notes

Video generation is VRAM-heavy

Text-to-video models typically need 8GB+ VRAM for comfortable local use. If you have less, use --service kaggle to offload to Kaggle's free T4 GPU.

Output videos are saved as .mp4 files in the specified output directory.