- Beam | AI Ads
- Posts
- Creating Winning Static Ad Images with ChatGPT
Creating Winning Static Ad Images with ChatGPT
How to get 95% of the way to production-ready ads with AI
Safwaan Khan
March 31, 2025
Table of Contents
Quick Intro
90% of the results, in 30 seconds
Use one product image for one generation
Minimize the amount of text that ChatGPT needs to generate (both on products and in the ads)
Give ChatGPT multiple images if you’d like it to use a specific product or a particular brand style
Specify the output size you’d like (1:1, 9:16, etc)
Give it an example of an ad you’d like to recreate. (You can use Beam.new. It has 25k+ curated ads, 1-click and paste to ChatGPT)
Generating consistent “on brand” designs
Attach 5-6 designs to ChatGPT that adhere to your brand style.
Use one of the following prompt to get your brand style
Brand Design & Style Guidelines Prompts
Option 1 (JSON)
Give me a detailed description of the visual identity of this brand. Output in JSON format.
Option 2 (No JSON)
Using the provided images as inspiration, develop a comprehensive document outlining the brand's design and style guidelines for ads and product shoots that captures its essence for anyone to recreate its look. Include a clear description of the brand’s identity—its personality, mission, core values, and the key messages or emotions it should evoke; an analysis of the mood and vibe conveyed (e.g., energetic, minimalistic, luxurious, playful) to define the visual tone; a detailed color palette listing primary, secondary, and accent colors with guidance on their use across mediums; specifications for typography and graphic elements, noting fonts, motifs, textures, or patterns that reflect the brand’s character; photography guidelines covering composition, lighting (natural vs. studio), angles, focus, props, backdrops, and shot types (close-ups, lifestyle images, flat lays); styling and composition rules addressing framing, spacing, layout, and dos and don’ts for image styling and product placement; an explanation of the intended audience reaction and how the visuals contribute to the overall brand experience; and any additional observations or nuances, including seasonal or campaign-specific adaptations, to ensure a consistent and distinctive visual identity
Once you get a response from ChatGPT, attach the ad / design you’d like to remix, and ask it to:
“Remake the following ad but for my brand, <BRAND_NAME>. Replace the visuals and copy in the attached image with ones that represent my brand.“
OPTIONAL: Add messaging and image guidance to the prompt above, to control the output. (For best results, use one product image ONLY).
Preventing product distortion
You have 2 options:
Simple & most reliable: Use your product in a vertical shot, with front facing the camera.
Advanced: Attach multiple images of the same product, from different angles. 5-6 should do the trick.
Then, attach an image of the ad / design you want to recreate. IDEALLY, this example image should have a product that is the same dimensions as your product. Example, a model holding a Coke can would work best if you have a canned beverage.
Use the following prompt (with additional guidance on messaging, branding, etc):
Replace the image in the given design with my image, attached. Do NOT distort the image. Make sure it obeys the laws of physics as it relates to orientiation, lighting, relative sizing, etc.
Making sure text doesn’t look like gibberish
Before generating your design, paste your product in ChatGPT (4o, NOT mini) with the following prompt.
Describe all the text elements in great detail, including their case, font sizes, font coloring, spacing, absolute, and relative locations. Output as JSON.
Then continue as usual in the same chat. Ask ChatGPT to create an ad with your product (or remix a different one). IMPORTANT: Copy the JSON output above and paste it into your next message. (Sometimes ChatGPT ignores the previous JSON output, despite being in the same chat.)
NOTE: The best option is to ask ChatGPT to create the product shot / images WITHOUT TEXT. Then add the text in post-production (Figma, Canva, Photoshop).
Resizing ads for 9:16, 1:1, 9:16, 4:5
The easiest way is to NOT resize the image. Instead, force ChatGPT to use the following safezone template. When Meta crops the image, it should look good on all formats.
More info here. Download the multi-format safezone template here → Multi-format-safezone.png
If you want the ads to be native to each format, do the following:
Create your AI Ad in 1:1 format
Open a new ChatGPT window
Use the following prompt
“Please remake this image in <SIZE> format. I’ve attached a safezone placeholder image for where you should NOT place any elements (in grey). When remaking the image, DO NOT CHANGE ANY IMAGES, TEXTS, FONTS, ETC. Keep everything the same as the original image, just resize it.
Safezone templates
Adhering to platform safe zones
See above
Bonus
Making sure your ads aren’t illegal
Use the following prompt with your ads attached.
You are a compliance and policy expert specializing in Meta’s (Facebook, Instagram) advertising guidelines and intellectual property law. Review the attached image as if submitted for a paid Facebook or Instagram ad. Assess it against Meta’s Ad Policies and legal standards, focusing on prohibited/restricted content (e.g., nudity, hate speech, violence, banned products), intellectual property use (e.g., logos, copyrighted material), deceptive claims, and politically or socially sensitive content. Determine if the ad is “Likely fine” (compliant) or “Likely not fine” (potential or clear violations). If concerns exist, specify what’s problematic, why, and how to fix it. Image to Review: [ATTACH IMAGES TO CHAT] Output Format: Verdict: Likely fine / Likely not fine Explanation: Brief summary of key issues Actionable Suggestions: Required fixes or confirm no changes needed.
Finding & clone winning ads
Visit beam.new
Search for your competitors (ex. Yeti), your industry (ex. Electronics), or format (ex. Before vs After).
Find an ad you like. Copy it. Click “Open in ChatGPT”
Paste it & recreate it for your brand (by attaching an image of your product and a prompt of what you’d like to change).
Creating photorealistic product photoshoots
Attach multiple 5-6 images of the same product, from different angles.
Then, attach an image of the product shoot you’d like to recreate. IDEALLY, this example image should have a product that is the same dimensions as your product. Example, a photoshoot from Burberry would work well if you also sell trench coats.
Use the following prompt (with additional guidance on messaging, branding, etc):
Recreate the following photoshoot but with my product. I’ve attached my product from different angles so that you can ensure it looks natural and exactly like my product. Do NOT distort the image. Make sure it obeys the laws of physics as it relates to orientiation, lighting, relative sizing, etc.
Turning static ads into video ads
Create an account on fal.ai
Choose either of the following options
$ → (1 cent per second) → fal-ai/kling-video/v1.6/pro/image-to-video
$$$$$ → (50 cents per second) → fal-ai/veo2/image-to-video
Attach your product image. Enter a simple 1-3 sentence prompt. It should take about ~5 minutes to generate a 5 second clip.
WTH is JSON format & why should I care?
JSON is just a way to format text for computers to read it easily. You don’t necessarily need to use JSON, but it can make your outputs better, especially if you aren’t great at prompting. (It forces ChatGPT to think about the task in a more structured way, leading to better outputs).