ChatGPT DALL·E 3 Prompts With Midjourney

Use ChatGPT to develop and refine an image idea, then adapt the prompt for Midjourney to generate variations in its own style. You will need access to both services; the key is to treat ChatGPT’s wording as a starting point, not a prompt that transfers settings or images between the tools.

What You’ll Need

Access to ChatGPT with DALL·E image generation available, if you want to create a visual reference there. You can also use ChatGPT just to draft prompts.
Access to Midjourney through its current available interface, such as the Midjourney website or Discord, depending on your account.
A clear idea of the subject, intended use, and preferred image format. Have rights to any reference image you plan to upload.

Step-by-Step

Describe the goal before asking for a prompt. Tell ChatGPT what the image should show, who it is for, the mood, setting, composition, and intended aspect ratio. Mention any elements that must be included or avoided. For example: “Create a prompt for a cinematic editorial image of a cyclist on a misty coastal road at dawn. Keep the cyclist small in the frame, use muted colors, and leave clear space on the left for a headline.” Specific visual direction is more useful than a string of style adjectives.
Ask ChatGPT for a clean image-generation brief. Request one concise prompt for DALL·E and a separate version adapted for Midjourney. Ask it to keep the subject and composition consistent while making the Midjourney version visually descriptive and easy to edit. You could add: “Do not invent parameters; put any optional Midjourney settings on a separate line.” This reduces the chance that a suggested setting is mistaken for a requirement.
Review the DALL·E version. If you want a visual reference, use the DALL·E prompt in ChatGPT and inspect the result. Check whether the subject, framing, lighting, and mood match your goal. If not, ask for a targeted revision, such as “make the camera angle lower” or “remove the extra people.” Do not assume the image-generation result will match every instruction exactly.
Translate the idea, not the whole conversation. Copy the concise Midjourney prompt into the Midjourney interface. Remove conversational phrases such as “please create” and keep the concrete visual details: subject, setting, composition, lighting, color, and medium or aesthetic where relevant. A prompt is not a guarantee that Midjourney will reproduce the DALL·E image or interpret every phrase identically.
Add Midjourney parameters only when needed. If you want a particular shape, use the current aspect-ratio parameter supported by Midjourney, such as --ar 3:2. Add other parameters only after checking the current Midjourney documentation, because supported options and behavior can change. Put parameters at the end of the prompt and avoid asking ChatGPT to guess them.
Generate a first batch and compare results. Look for the version that best preserves your central idea, not necessarily the one with the most detail. If the results are too busy, simplify the prompt. If the subject is missing or unclear, move it to the beginning and describe its pose or defining features plainly.
Refine one variable at a time. Keep the strongest prompt as your baseline, then change just one aspect per attempt: composition, lighting, color palette, background, or level of detail. This makes it easier to see what caused an improvement. Save useful prompt versions and note any settings used so you can repeat a successful approach.
Use image references thoughtfully. If you want Midjourney to take visual cues from a DALL·E output, first confirm that your account and current interface support image prompts. Upload or select the image as instructed there, then pair it with a short text prompt explaining what to retain or change. Image-reference controls can affect results, but they do not promise an exact copy. Do not upload material you do not have permission to use.
Check the final image for practical issues. Inspect details such as hands, text, logos, faces, and background objects. Image models may render lettering or fine details inaccurately. If the image is for a project, confirm that its intended use complies with the services’ current terms and any requirements from your client or publisher.

Common Mistakes

Using one long prompt everywhere: ChatGPT may produce explanatory prose, while Midjourney needs a focused visual description. Copy the useful image details, not the surrounding instructions or discussion.
Expecting identical results: DALL·E and Midjourney are separate systems. The same words can produce different compositions, aesthetics, and details.
Stacking contradictory directions: “Minimal and highly detailed” or “bright sunlight in a dark night scene” can confuse the visual goal. Decide which quality takes priority.
Adding unverified settings: Parameters suggested from memory may be unsupported or behave differently now. Check current documentation before relying on them.
Changing too many things at once: If the prompt, aspect ratio, and reference image all change together, it is difficult to identify what helped.
Assuming generated text is reliable: Check any visible lettering manually; do not rely on an image model for exact wording, labels, or factual diagrams.

Tips

Put the main subject and action early in the prompt, followed by setting, framing, light, and color.
Use observable descriptions. “A red raincoat, wet pavement, and fog behind the subject” is clearer than “a mysterious, emotional atmosphere.”
Ask ChatGPT to provide a compact prompt plus a short list of optional details. This makes it easy to test additions without rewriting everything.
When adapting an existing DALL·E result, describe what to preserve and what to change separately. For example, preserve the camera angle but change the time of day.
Keep a simple prompt log with the text, reference image, and settings for each useful result. It helps you iterate without relying on memory.

Troubleshooting

Midjourney’s output is unrelated to the idea: Put the subject and action first, remove abstract or conflicting wording, and make the setting explicit. Try a fresh generation with the shorter version.

The composition is right but the style is wrong: Keep the subject and layout fixed, then replace broad style labels with specific cues such as medium, palette, lighting, or texture. Change one cue at a time.

The prompt is rejected or a parameter has no effect: Remove the questionable parameter and try the text alone. Check Midjourney’s current prompt and parameter guidance, as accepted options may change.

An image reference is not being used as expected: Confirm that it uploaded successfully and that you used the reference method supported by the interface. Add a concise instruction about which features matter most; do not expect an exact reproduction.

ChatGPT keeps returning vague prompts: Give it concrete constraints and ask for a prompt that describes visible details, not just mood. You can also paste a draft and request a shorter version without adding new concepts.

Frequently Asked Questions

Can I paste a DALL·E prompt directly into Midjourney?
Yes, but it may work better after removing conversational wording and adapting any settings. The two systems interpret prompts differently.

Do I need to generate an image with DALL·E first?
No. ChatGPT can help write or revise a Midjourney prompt without generating a reference image.

Can Midjourney recreate a DALL·E image exactly?
No. A reference may influence the result, but exact copying is not guaranteed and the tools can produce different details.

Can ChatGPT add Midjourney parameters for me?
It can suggest possibilities, but verify them against current Midjourney documentation before use.

How can I get more consistent results?
Keep a baseline prompt, change one detail at a time, and record useful wording and settings. Consistency is not guaranteed between generations.

Can I use the resulting image commercially?
Check the current terms for both services and any rules that apply to your project. Usage rights and conditions can depend on the service and account.

General steps for current mainstream versions; menus can differ by device and version. See the disclaimer or report an error.

More in this bed

The whole Tech How-To bed →