Midjourney Not Following Prompts Anymore? How to Fix It

If Midjourney has suddenly started ignoring your prompt details, dropping key elements, or prioritizing aesthetics over your written instructions, you are experiencing prompt drift. This behavior typically happens when model updates shift how tokens are parsed, or when style parameters overpower your text input.

Because newer models parse language differently than older versions, old prompting habits can cause the model to ignore your instructions. Follow these structured steps to diagnose and fix prompt adherence issues.

1. Shift from keyword lists to natural language Older Midjourney models (v4 and v5) relied heavily on comma-separated keywords (e.g., `photorealistic, cinematic lighting, ultra-detailed`). In newer models like v6, this "word salad" approach actively hurts prompt adherence. The model attempts to resolve conflicting stylistic buzzwords, which dilutes the actual subject of your prompt.

1. Write in full sentences: Describe the scene as if explaining it to a person. Instead of red car, rainy street, night, neon lights, write: A photograph of a wet red sports car parked on a city street at night under neon signs.
2. Remove empty descriptors: Eliminate words like *photorealistic*, *hyperrealistic*, *4K*, or *masterpiece*. Midjourney automatically applies high-quality aesthetics; these words only waste valuable token space.
3. Place critical subjects first: Midjourney reads from left to right. The first 5 to 10 words receive the highest attention weight. Place your primary subject at the absolute beginning of the prompt.

2. Reduce the Stylize parameter or use Style Raw By default, Midjourney applies an aesthetic bias to make images look artistic. If this bias is too strong, the model will ignore specific, unusual, or highly technical prompts in favor of a generic "pretty" image.

1. Use the Style Raw parameter: Append --style raw to the end of your prompt. This lowers the default aesthetic influence, forcing the model to adhere much more closely to your literal text instructions. 2. Lower the Stylize weight: The default stylize value is 100. If your instructions are being ignored, manually lower it by adding --s 50 or --s 0 to your prompt. For example: /imagine prompt: A square blue apple sitting on a gold plate --style raw --s 50

3. Apply Multi-Prompting and Text Weights When a prompt contains multiple distinct elements, Midjourney sometimes blends them together (color bleeding) or ignores one element entirely. You can force the model to process elements separately by using double colons (`::`) to assign mathematical weights.

1. Isolate the ignored element: If your prompt is a man wearing a yellow hat eating a green apple, and the apple keeps turning yellow, break it up. 2. Assign weights: Use :: to separate the concepts and assign a higher number to the ignored object. *Example:* a man wearing a yellow hat :: eating a green apple ::1.5 3. Ensure weights balance out: Do not assign massive weights (like ::5) as this will distort the image. Stick to small adjustments between 0.5 and 2.0.

4. Lower Reference Image Weights Using Character References (`--cref`) or Style References (`--sref`) can completely overpower your text prompt, causing Midjourney to ignore new instructions in favor of matching the reference image.

1. Lower Character Weight: If using --cref, the default character weight is 100, which copies clothing, hair, and face. If you want to change the character's clothes or action and the model is ignoring you, add --cw 0. This forces the model to only copy the face, giving your text prompt control over the rest of the scene.
2. Lower Style Weight: For style references, use the --sw parameter to scale back the style's strength. The range is 0 to 1000 (default 100). Try setting --sw 25 or --sw 50 to let your text prompt override the reference style.

5. Verify the Active Model Version If your prompts are suddenly yielding completely different, lower-quality, or inaccurate results, your settings may have reverted to an older model version.

1. Type /settings in your Discord server or prompt bar and press Enter.
2. Look at the active model buttons. Ensure Midjourney Model V6 (or the latest version) is selected.
3. If an older version like v5.2 or Niji is active, click the V6 button to restore modern text parsing.

When to escalate If you have simplified your prompt, applied `--style raw`, set `--s 50`, and verified you are on V6, but the model still completely ignores basic nouns or behaves erratically, there may be a system-wide bug. Check the official Midjourney Discord in the `#member-support` or `#status` channels to see if there is an ongoing server-side issue with token parsing or model deployment.