# Why I built an open creative layer across image, video, voice and AI agents

> Source: <https://dev.to/prantik_1995/why-i-built-an-open-creative-layer-across-image-video-voice-and-ai-agents-5cke>
> Published: 2026-09-14 06:18:17+00:00

Foundation models keep changing. Your creative workflow shouldn't have to.

AI media has an unusual problem.

The best model for images may come from one company.

The best model for video may come from another.

Your preferred text-to-speech model may be somewhere else entirely.

And six months later, all three might change.

That makes building a creative application surprisingly awkward.

If your product is deeply coupled to today's winning model, you're effectively building on shifting sand.

While building OpenCreative, I started approaching the problem differently:

Treat models as infrastructure. Treat the creative workflow as the product.

Then continue with sections:

Explain:

users lose project context between products;

assets have to be downloaded/re-uploaded;

prompts and references don't travel;

pricing/model quality changes constantly.

Explain that the user asks to:

generate an image;

animate a frame;

dub a video;

create a voiceover;

relight footage;

rather than having to think first about which provider should perform it.

Example:

A user starts with a product launch brief.

Generate product photography.

Turn the hero shot into video.

Add narration.

Translate it.

Produce social cuts.

Reuse the brand material for a customer-facing support agent.

That's the workflow OpenCreative is trying to make continuous.

Talk about:

avoiding permanent vendor lock-in;

provider abstraction;

future BYOK;

self-hosting where practical;

keeping private assets under stronger user control.

Good points:

abstractions should expose advanced controls instead of hiding everything;

creative consistency matters more than raw model count;

cost estimation should happen before generation;

asynchronous media generation changes UX considerably;

storing lineage between input/reference/model/output is extremely useful.

Closing

I'm still early with this.

OpenCreative currently includes image, video, voice, dubbing, music, avatar/character and agent workflows, but I'm much more interested in whether the architecture survives as models change.

I'd especially like feedback from people building on multiple AI providers.

Project: [opencreativehq.com](https://opencreativehq.com/)

If there's interest, I can write a second post about how we're handling provider routing, asset lineage and generation jobs.

That last sentence gives you an excuse to publish Part 2, which means more backlinks and more discovery.