Describe the work.
The agent builds it.
Most AI tools turn one prompt into one output, and leave the thinking to you. The Creative Agent takes a brief, plans the work, chooses a model for each step, and iterates until the result matches your brand.
Prompting a model
- You engineer the prompt
- One prompt, one output
- You restate brand rules every time
- You sort the batch and retry the misses
Creative Agent
- You describe the outcome
- A brief becomes a plan with steps
- Brand rules are already in context
- It reviews its own work and re-runs misses
What it actually does
Plans before it generates
The agent decomposes a brief into concrete steps and shows you the plan. You see what it intends to make before credits are spent.
Picks the right model per step
Different models win at different jobs. The agent routes each step to a suitable one instead of forcing everything through a single provider.
Reviews and iterates
It checks its own output against the brief and your brand rules, then regenerates what misses — rather than handing you a batch to sort through.
Brand-aware by default
Your Brand OS profile is part of the agent context. Colours, fonts, logo rules and tone apply without being restated in every brief.
Calls your workflows
Existing automations become single steps the agent can invoke, keeping repeatable production deterministic.
Steerable mid-run
Redirect it, correct a step, or pin an asset you want kept. Nothing is locked in until you accept it.
How it works
Describe the work
A brief in plain language — no prompt engineering
Review the plan
See the steps and assets before anything runs
Agent produces
Generates, checks against brand, iterates on misses
Take it further
Accept, refine in Canvas, or push into a workflow
Questions
What is the Creative Agent?
The Creative Agent takes a brief in plain language and works it through to finished creative. Rather than returning a single generation from a single prompt, it breaks the brief into steps, picks the right model for each one, generates, reviews its own output, and iterates until the work matches what you asked for.
How is this different from prompting an image model directly?
A prompt returns one output. An agent runs a process. It decides how many variations a brief needs, which model suits each asset, what your brand rules require, and what to change when a result misses. You describe the outcome instead of engineering the prompt.
Does it know my brand?
Yes. The agent reads your Brand OS profile — logos, colour palette, typography, tone of voice and visual references — and applies it to everything it produces. You do not have to restate your brand rules in the brief.
Which models can it use?
All of them. The agent has access to every image, video, audio and language model available in Pixloop, and chooses per step rather than locking you to one provider.
Can I intervene while it works?
At any point. You can redirect the agent mid-run, correct a step, pin an asset you want kept, or take a result into the Canvas and finish it by hand. It is a collaborator, not a black box.
Can it run an existing workflow?
Yes. If you have already built a workflow, the agent can call it as a single step — so repeatable production stays deterministic while the agent handles the parts that need judgement.