Revolutionizing Ads

• 

  workflow-video-generation

• 

  PixVerse-Video-Editing

• 

  Architecture diagram

Inspiration

Creating video ads is still slow and rigid. Teams often commit to a single video model early, making it hard to iterate, compare styles, or make fast edits. We wanted to make ads modular, editable, and powered by the best of multiple AI models.

What it does

We built a multi-model AI video ad pipeline that:

Starts from one script

Generates multiple videos using different text-to-video models

Lets creators easily edit and refine results using PixVerse

This makes it simple to compare outputs, remix scenes, and update ads without regenerating everything.

How we built it

Generated a unified script using OpenAI OSS 120B

Sent the script in parallel to:

WAN2.6

Veo3

PixVerse 5.5 (text-to-video)

Ingested all videos into PixVerse Edit, using its APIs for masking and targeted edits to refine and unify the final ad

PixVerse acts as the editing and control layer across all models.

Challenges we ran into

Aligning outputs from different video models

Preserving creative consistency across styles

Enabling fast edits without full regeneration

Accomplishments that we're proud of

A model-agnostic ad creation workflow

Fast post-generation editing with PixVerse

Side-by-side comparison of multiple AI video outputs

A pipeline that matches real ad iteration workflows

What we learned

No single model is best at everything

Editing is just as important as generation

Modular, multi-model pipelines unlock better creativity

What's next for Revolutionizing Ads

Automated scene ranking

Brand-aware edits

Prompt-based video editing

A/B testing for ad performance

Built With

• gmi-cloud
• llms
• pixverse
• veo3
• wan