Multi-Model Creative Campaign Pipeline
Pair a reasoning model with an image model to build coherent campaign assets
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 98%
Start by assembling a detailed creative guide containing the brand's visual system, voice, audience, product, and campaign objective. Give that context to a strong reasoning model and ask it to act as a creative director, research the product, and write a master prompt. The prompt should sequence the work: concise creative brief, campaign concept, mood board, billboard, and format-specific digital ads with dimensions and output requirements. Transfer that prompt to an image-capable model, which generates the brief and visual assets. Because image generation may return only one asset at a time, call back to the same established instructions and request each remaining format individually. Review the outputs for campaign coherence, accurate claims, legible copy, brand adherence, and production readiness before use.
Origin
Extracted from Marketing Against the Grain through a HubSpot Customer Agent campaign demonstration using o3 for creative direction and GPT-4o for image generation.
Core principles
- 01Assign each model the task that matches its strengths
- 02Ground every output in a complete creative guide
- 03Research the product before developing the campaign
- 04Generate strategy before generating visual assets
- 05Maintain one shared brief across every campaign format
- 06Request expensive image assets deliberately and iteratively
How to run it
- 1
Assemble the creative guide
Collect the brand's approved fonts, colors, illustration style, photography conventions, tone, and public visual references. Include enough detail for a model to distinguish the brand from generic category creative.
Pro tip Use a researched style guide rather than a short list of colors.
Watch out Do not include confidential material unless the chosen system is approved to process it.
- 2
Define the campaign
State the target audience, promoted product, core objective, required asset formats, and intended placements. Ask for product research when the model needs factual grounding.
Pro tip Specify image dimensions and output formats in the initial request.
Watch out Unverified product claims can migrate into every generated asset.
- 3
Direct the reasoning model
Have a capable reasoning model act as the creative director and write a master prompt for the image-capable model. Require an ordered workflow rather than a loose creative request.
Pro tip Ask it to express the instructions in structured markup if that improves compliance.
Watch out Do not ask the image model to invent the entire strategy from a one-line prompt.
- 4
Create the brief and mood board
Generate a concise creative brief with the proposition, benefits, campaign idea, and visual direction. Produce a mood board that establishes a consistent visual vocabulary before final assets.
Pro tip Check whether the mood board correctly reflects the supplied brand system.
Watch out A visually attractive mood board is not sufficient if it contradicts brand rules.
- 5
Generate each campaign asset
Request the billboard, square ad, and other formats from the same conversation and master prompt. If the model omits an asset, explicitly call back to the existing instructions and request it separately.
Pro tip Keep all requests in the same context to preserve consistency.
Watch out Expecting many expensive images from one request may lead to incomplete output.
- 6
Review and refine
Check legibility, hierarchy, calls to action, URLs, product claims, dimensions, and visual consistency. Correct weak assets individually without discarding the campaign's shared direction.
Pro tip Evaluate the work in its intended placement, not only as a standalone image.
Watch out Never publish generated creative without checking factual claims and usage rights.
In the wild
The host gave o3 HubSpot's creative guide and asked it to prepare a prompt covering research, a creative brief, a mood board, a billboard, and a square digital ad. He moved the resulting prompt into GPT-4o, which developed the campaign proposition and generated branded assets. When the model omitted requested images, he prompted for the mood board and digital ad separately while retaining the original campaign context.
→ A coherent initial campaign with strategy, copy, visual direction, a billboard, and a digital advertisement.
Common mistakes
Using one model for every task
A model skilled at strategic reasoning may not be the best image generator. Route creative direction and visual production to models suited to each job.
Requesting visuals without a shared brief
Independent image prompts produce inconsistent copy and art direction. Establish one creative brief and reuse it across all formats.
Assuming every asset was generated
Image models may return only one requested item. Check the deliverables and explicitly request missing assets from the existing context.
Is it for you?
Best for
It is best for marketers producing an initial campaign concept, mood board, billboard, and digital advertisements from an established brand guide.
Not ideal for
It is not ideal for final regulated or high-budget creative that cannot receive human brand, legal, and production review.
From the transcript
“I think one of the kind of master class parts of using AI right now is using different models to accomplish a task.”
“Hey, first you're going to need to research, then you're going to write a concise creative brief, generate a campaign, mood board image, Billboard concept,…”
“it's kind of a one image at a time situation, but I still like to go through and put them all in the prompts, because…”
From the episode
ChatGPT o3 Deep Dive: 3 Wild New Use Cases