Automate Midjourney Image Prompts with ChatGPT
The Bottleneck of Manual Image Prompting
I used to waste ninety minutes every morning in my Seoul studio manually copying color palettes, aspect ratios, and camera angles into Midjourney. Running multiple content blogs and video channels requires dozens of featured images, thumbnail concepts, and inline diagrams every week. Doing that hand-by-hand meant I was spending more time engineering text prompts than actually producing content.
The issue is not coming up with visual ideas. The issue is translating a blog post concept into the specific syntax that Midjourney expects. You need lighting styles, artistic genres, composition details, and parameter flags like --ar 16:9 or --v 6.0. When you try to generate Midjourney prompts with ChatGPT directly without a clear structure, you end up with generic fluff like ‘a beautiful realistic photorealistic image of a laptop.’ That output produces poor Midjourney renders and wastes your GPU fast hours.
To fix this, I set up an automated prompting pipeline that turns simple article titles or rough outlines into production-ready Midjourney prompt strings. Here is how I built the system, where it fails, and how you can replicate it for your own publishing workflow.
Step 1: Constructing the ChatGPT Prompt Architect
Standard ChatGPT responses are too conversational for image generation tools. Midjourney performs best with concise, comma-separated keywords and explicit parameter tags at the end of the string. To force ChatGPT to output clean prompts, you must define a custom system prompt.
When I created my custom prompt generator inside ChatGPT, I established strict rules for how it formats visual ideas. You can paste the following system instruction directly into a standard ChatGPT chat or build a Custom GPT with it:
- Role: Act as an expert Midjourney Prompt Engineer.
- Task: Convert raw article concepts, topics, or headings into precise Midjourney image prompts.
- Structure: [Primary Subject], [Environment/Background], [Lighting/Mood], [Art Style/Medium], [Camera/Composition Details] –ar [Aspect Ratio] –v 6.0 –stylize [Value].
- Constraints: Do not use full conversational sentences. Avoid buzzwords like ‘photorealistic’ or ‘hyperrealistic’ (Midjourney v6 handles realism through camera details like ’35mm shot, f/1.8 lens’). Never include introductory filler text like ‘Here is your prompt:’. Output only the raw prompt.
The mistake I made early on was letting ChatGPT pick random aspect ratios. I ended up with square images for YouTube thumbnails and landscape images for Pinterest pins. Now, I explicitly instruct ChatGPT to default to --ar 16:9 for web articles and YouTube, or --ar 9:16 for social reels.
Step 2: Automating Bulk Prompt Generation
Generating prompts one by one still requires tab-switching. If you publish five articles a day, you need a bulk process. I handle this by feeding ChatGPT a structured list of blog post headers and requesting a batch table output.
When you provide a topic list, instruct ChatGPT to return a Markdown table with three precise columns: Section Title, Visual Concept, and Final Midjourney Prompt. Here is a typical workflow table produced by this method:
| Section Title | Visual Concept | Midjourney Prompt |
|---|---|---|
| AI Automation | Robot workspace | Minimalist desk, robotic arm typing on mechanical keyboard, warm ambient lamp lighting, editorial photography, cinematic, 50mm lens –ar 16:9 –v 6.0 |
| Data Security | Digital shield | Abstract glass shield icon glowing blue over server racks, corporate office background, dark mood, isometric render, octane render –ar 16:9 –v 6.0 |
| SEO Strategy | Search growth | Surreal glass ladder reaching into cloud-filled sky, golden hour light, soft focus, architectural photography, minimal design –ar 16:9 –v 6.0 |
By enforcing a table format, you can copy the entire column directly into your clipboard or export it to a spreadsheet. This simple shift cut my prompt creation time down from hours to roughly two minutes per site.
Step 3: Connecting ChatGPT to Your Automation Pipeline
If you want to go beyond copy-pasting, you can integrate ChatGPT into an automated workflow using platforms like Make, Zapier, or n8n. In my setup, when a new article draft is added to Notion or WordPress, an automation hook triggers an OpenAI API call.
The API request sends the article summary to OpenAI using the custom system prompt described in Step 1. The API returns the completed Midjourney prompt and writes it back into a designated Notion property or custom field in WordPress.
However, you need to keep Midjourney’s official policies in mind. Midjourney currently does not offer an official public REST API for direct image generation. While third-party automation tools exist to send text prompts directly into Discord, using unauthorized Discord bots can put your Midjourney account at risk of suspension. For my channels, I automate the ChatGPT prompt generation completely via API, but keep the final step—pasting the prompt string into Midjourney’s web interface or official Discord—as a quick manual step or semi-automated trigger.
Common Pitfalls and How to Avoid Them
When I first automated this system, several things went wrong. Watching out for these common errors will save you time and generation credits:
- Over-describing the scene: ChatGPT loves long prose. Midjourney v6 prefers shorter keyword clusters. If your prompt exceeds 60 words, Midjourney often ignores the parameters at the very end. Keep ChatGPT’s character limit tight in your instructions.
- Hallucinating parameter syntax: Early on, ChatGPT kept inventing parameters that do not exist, like
--quality 5or--resolution 4k. Ensure your prompt generator system instructions strictly define allowed flags (e.g.,--ar,--v,--chaos,--stylize). - Neglecting subject variety: If you ask ChatGPT for five prompts about ‘SEO software’, it will give you five pictures of laptops on modern desks. Instruct ChatGPT to vary the style across categories like isometric 3D renders, vintage editorial photography, oil paintings, or minimal vector art.
My Take
Using ChatGPT to generate Midjourney prompts is one of the highest-ROI workflow tweaks a solo operator can make. It strips away the friction of staring at a blank cursor trying to decide on camera lenses or lighting modifiers.
That said, total ‘hands-off’ visual automation is a myth if you care about visual quality. ChatGPT gives you a solid 90% starting point, but you will still need to tweak parameters, swap out styles, or run variations on Midjourney to get a final image that matches your brand tone. Treat ChatGPT as your visual creative director, not an unsupervised publisher.
Regarding costs: ChatGPT offers a free plan, while paid tiers with custom GPT access start around the $20/month range. Midjourney basic subscription tiers start around $10/month range. Always check official pricing pages for OpenAI and Midjourney as subscription models update frequently.
FAQ
How do I make ChatGPT output Midjourney parameters like –ar 16:9?
You must explicitly instruct ChatGPT in your prompt or system settings to append parameter strings at the end of every line. Provide concrete examples in your initial prompt showing the exact syntax you expect (e.g., ‘Ending format: [description] –ar 16:9 –v 6.0’). If ChatGPT omits them, add a negative constraint stating ‘Do not output prompts without aspect ratio flags.’
Is it better to use Custom GPTs or the standard ChatGPT chat for prompts?
Building a Custom GPT is better if you regularly generate prompts. With a Custom GPT, you save your formatting rules, preferred aspect ratios, and parameter defaults permanently into the GPT instructions. You simply drop in an article text, and it instantly outputs formatted prompts without needing background instructions every session.
Can ChatGPT generate Midjourney prompts in bulk from an article draft?
Yes. You can paste an entire article draft into ChatGPT and instruct it to identify the three key visual themes, then generate one Midjourney prompt for each theme formatted as a list or table. This allows you to generate feature images, mid-post diagrams, and social graphics in a single request.
