AI Creative Engines: Auto-Generating Winning Ad Variants with GoMarble, Weavee, and DALL-E
What is GoMarble and how does it optimize ad campaigns?
GoMarble is a performance marketing optimization engine that connects to your ad campaigns through APIs and inspects both the campaigns and the assets attached to them.
From there, it recommends how each campaign should be optimized — either on the text or on the creative .
It sits downstream of your measurement layer.
If you have unified marketing modeling in place — multi-touch attribution and marketing mix modeling done correctly and combined — GoMarble is optimizing against conversions you actually trust.
Worst case, you're at least feeding it last-touch data correctly.
The tool can reportedly be white-labeled, and the workflow is API-first: campaign and asset data flows into a central engine that returns optimization recommendations, removing the human from combing through Excel sheets and campaign reports to identify winners.
How do I auto-generate ad creative from winning concepts?
Once an engine like GoMarble surfaces which campaigns and assets are winning, the next step is producing more of them automatically.
The pattern described by practitioners is to chain the optimization engine into an image generation stack: Weavee — an image orchestrator tool used to create image generation workflows.
DALL-E 3 — an image generation LLM that produces the visuals themselves.
With that chain in place, you can auto-generate visuals from winning concepts .
The explicit point of view: a human shouldn't be doing this.
Humans shouldn't be going through Excel sheets or campaign reports to find the most successful assets and then manually briefing new versions — the orchestration is the job.
This matters because the paid social algorithms themselves have shifted.
Meta is increasingly telling advertisers to hand over objective, product info and creative, and let the platform find the buyer — so the work moves off the keyboard and onto feeding the algorithm volume and variety of creative to test.
“The more different types of creative you can feed the AI, the more it can test out different things and find combinations of audiences and messaging.”
Why does this creative-engine stack matter right now?
Media buying is something AI can do very well, and the trend on Meta and Google is unmistakable: less time on keyboard manually manipulating the algorithm, and more time feeding it data — specifically creative.
As one operator put it, the AI is finding customers based on reactions to content .
That reframes the marketer's job.
The competitive edge moves to teams who can churn out lots of creative — creative strategists, people connected to affiliate marketers and influencers, and people who can generate huge amounts of unique creative with unique hooks and high production value .
That's the gap a GoMarble + Weavee + DALL-E 3 stack is built to close: optimization tells you what's winning, orchestration plus generation produces more of it, and the platform algorithms do the targeting.
Frequently asked questions.
- What is GoMarble?
- GoMarble is a performance marketing optimization engine. It connects to your ad campaigns, inspects the campaigns and the assets associated with them, and then makes recommendations for how each campaign should be optimized — either on the text or on the creative side. It can reportedly be white-labeled, and it's designed to replace the manual work of combing through Excel sheets or campaign reports to find your best-performing assets.
- What does Weavee do in an AI ad creative workflow?
- Weavee is an image orchestrator tool used to create image generation workflows. In a typical stack, it sits between your optimization engine and your image generation model — for example, taking winning concepts surfaced by a tool like GoMarble and routing them through an image generation LLM such as DALL-E 3 to produce new visual variants automatically.
- How does DALL-E 3 fit into paid social creative production?
- DALL-E 3 is described as an image generation LLM — meaning it's the layer that actually produces visuals on demand. Inside a creative engine workflow, it's chained behind an orchestrator like Weavee so that winning concepts from your performance campaigns can be auto-generated as new visuals, rather than briefed manually each time.
- Why are platforms like Meta pushing advertisers to hand over creative and let the algorithm run?
- Meta is increasingly telling advertisers to specify their objective, share product information and feed in ad creative, and then let the platform find the buyer. This is correlated with AIs getting better — the algorithm finds customers based on their reactions to content. So advertisers spend less time on keyboard manipulating the algorithm and more time feeding it new data, especially creative variants.
- Why does volume and variety of creative matter so much now?
- Because paid social algorithms are now finding customers based on reactions to creative, the more different types of creative you can feed them, the more combinations of audiences and messaging they can test to find winners. The competitive edge is shifting to teams who can generate huge amounts of unique creative with unique hooks and high production value — which is exactly what an auto-generation stack enables.
- How does this connect to attribution and measurement?
- Optimization engines only help if the conversions they optimize toward are real. The most sophisticated setups combine multi-touch attribution done correctly with marketing mix modeling done correctly — together called unified marketing modeling — so the optimization engine is acting on trustworthy signals. At minimum, you want last-touch tracked correctly before piping campaign data through APIs into a central optimization engine.