OpenAI says Pollo Agent turns concepts and references into scripts, storyboards, images and short ads by assigning different models to each task.
Pollo AI’s Pollo Agent uses GPT‑5.6, GPT‑6 Astra and GPT‑Image‑2.5 to turn rough ideas and visual references into storylines, scripts, scene sequences and finished campaign assets, according to an OpenAI profile published on October 8. The product is aimed at creators and marketers producing images, videos and advertising materials.
Each model has a distinct role in the workflow. OpenAI says GPT‑5.6 routes requests, while GPT‑6 Astra handles more demanding narrative development and plans revisions to individual scenes. Pollo AI reports that this allocation reduces the time users spend selecting or switching models by more than 50%, although the source provides no independent test of that figure.
GPT‑Image‑2.5 generates posters, storyboards and other campaign visuals. According to OpenAI, Pollo AI assesses candidate models for quality, visual consistency, stylistic range, speed and cost. The company says its service has more than 26 million users and that 30% of creator sessions begin with a template or workflow based on a trending format.
One featured workflow is Photo to Video Ads within Marketing Studio. A user supplies a product photograph and a prompt, and the tool assembles a short advertisement for social media. In OpenAI’s fragrance example, the system produced a video lasting about 15 seconds, with generated textures, lighting, close-up shots and a staged final composition.
Other showcased concepts include a jewelry campaign placing a model beside a cheetah and a beverage advertisement that turns a bottle photograph into a miniature snowy fantasy landscape. OpenAI attributes these workflows to Pollo Agent and GPT‑Image‑2.5, saying they can reduce production effort, time or cost compared with live shoots. These are publisher-presented examples rather than independently assessed results.
The practical proposition is a single workflow that assigns models without requiring users to choose a separate generator at every stage. However, the source gives no pricing or detailed access terms, and it does not report independent evaluations, latency measurements, accuracy rates, safety results or quantified comparisons of output quality. Claims about savings and effectiveness should therefore be read as company-reported performance, not as an industry benchmark.
Sources
Originally published by AI Feed: Pollo Agent Uses Three OpenAI Models to Build Ad Campaigns.
Top comments (0)