How to Use --iw (Image Weight) in Midjourney to Actually Control Your Output
When you add an image URL to a Midjourney prompt, the model uses it as a visual reference alongside your text. How much influence that image has relative to your text prompt is controlled by the --iw parameter. Without understanding this parameter, you are guessing at a number that fundamentally determines whether your output looks like the reference, like your text prompt, or somewhere between the two.
This article explains exactly what --iw does, how to set it correctly for different use cases, and the common mistakes that make image prompting feel unpredictable.
What --iw Actually Controls
When you provide an image URL in a Midjourney prompt, the model processes both the image and the text and blends their influence on the final output. --iw sets the relative weight of the image versus the text in this blend.
- At low
--iwvalues, the text prompt is dominant. The image provides light influence on color or mood but your text description drives the content and composition. - At high
--iwvalues, the image is dominant. The output will more closely resemble the reference in composition, color, subject matter, and overall visual feel. Your text prompt becomes a modifier of the image rather than the primary driver.
The valid range for --iw in Midjourney V6 and V7 is 0 to 3. The default value is 1 when an image prompt is used without specifying --iw.
Understanding the Value Range in Practice
--iw 0 to 0.5: Image as loose color reference
At very low values, the image has minimal impact. You might see subtle color temperature alignment or a vague compositional echo, but the text prompt dominates almost entirely. Useful when you want an image to hint at a mood or palette without pulling the output toward the reference subject.
--iw 0.5 to 1.0: Soft influence, text-led
The default range. The image provides a meaningful aesthetic pull without overriding the text. Useful when you have a reference image that establishes a general feeling or style but you want the text to define the content clearly. This is a good starting point for style transfer work.
--iw 1.0 to 1.5: Balanced blend
Image and text carry roughly equal weight. The output will feel like a synthesis of both. Compositions from the image may appear, and the text description shapes subject matter and detail. Most useful when your reference and text are both important and you want neither to dominate.
--iw 1.5 to 2.5: Image-led output
The image becomes the primary driver. Composition, subject matter, and color are pulled strongly toward the reference. Text prompt becomes a modifier, adding elements or adjusting mood rather than defining the scene. Useful when you are iterating on an existing image, applying a specific visual style, or need the output to closely match a reference composition.
--iw 2.5 to 3.0: Near-direct image interpretation
At maximum values, the output is very close to the reference image with text modifications layered on top. Use with care: at these values, text prompts that conflict with the reference are largely ignored. Useful for variations on a specific existing image rather than creative exploration.
Three Specific Use Cases With Recommended Settings
Style transfer: applying a visual aesthetic to a new scene
Goal: capture the color, rendering style, and mood of a reference image while generating entirely different content.
Recommended: --iw 0.75 to 1.25. At this range, the style characteristics carry across without forcing the composition or subject of the reference into the output. If you want stronger style consistency at the cost of some content freedom, move toward 1.5.
Note: for pure style transfer without content influence, --sref is often a better tool than image prompting with --iw, because --sref is specifically designed to extract aesthetic properties rather than compositional ones.
Composition anchoring: using a reference to guide layout
Goal: use an existing composition structure (where the horizon sits, subject placement, depth layers) as a guide for a new scene with different content.
Recommended: --iw 1.5 to 2.0. At this range, the compositional structure of the reference has a strong influence while the text prompt defines what fills that structure. Works best when the reference and target content have compatible subject scales (both landscape, both portrait figure, etc.).
Variation generation: producing alternatives to an existing image
Goal: create new images that are clearly related to an existing output without being identical.
Recommended: --iw 1.75 to 2.5. At these values, the output shares strong visual DNA with the reference. Combined with varied text prompts, you can produce a family of related images rather than random variations. Note that Midjourney's built-in Vary tools (Subtle and Strong) may be more efficient for this specific use case.
