Home Creators Posts Import Register Favorites Logout

Content

One of the oldest sliders in Midjourney just got an overhaul. In v8.1, image weight (--iw) has been slightly overhauled in ways that make it more precise, more punishing at extremes, and more useful in the middle range than any prior version

What --iw Does (The Short Version)

Image weight controls the tension between your reference image and your text prompt. Low values = the text wins. High values = the image wins. In v8.1, that gradient is smoother and more predictable than any previous version.

Range is 0 to 3. Default is 1.

The Reference Image

For this experiment, I am using the famous Francis D. by Andrei Kovalev, which some of you are familiar. This is a face with texture, age, and character. The lighting is dramatic studio portraiture against a near-black background.

The prompt used is "fancy photoshoot with this man --seed 0001" – I want the picture have some influence when weight gets highter.

Value-by-Value Breakdown

--iw 0: Text Takes Full Control

The model essentially ignores the reference entirely – it's like you never put a reference at all. What appears is a young Black man in opulent baroque clothing — crimson-and-gold brocade, ruffled shirt, jeweled ring — seated on a velvet throne. The text prompt's "fancy photoshoot" interpretation dominates completely. The reference subject has vanished.

--iw 0.25: A Ghost of the Reference

Still dominated by the text. Another young Black man, similar baroque styling, slightly different pose and build. The reference's facial character has no foothold yet. The model is still essentially doing free-form generation from the text alone.

--iw 0.5: First Signs of Identity Transfer

Something subtly shifts — the face reads as slightly older, the posture more contemplative, the expression heavier. The reference is beginning to influence mood and weight, even if not literal appearance.

This is where emotional resonance starts bleeding through before physical likeness does.

--iw 0.75: Character Starts to Arise

The figure is now distinctly thinner, more angular. The face has acquired pronounced cheekbones, a leaner jaw, a more intense gaze. The baroque costume continues, but the body language is quieter, more interior. The reference's lean, weathered quality is transmitting — just into a different demographic shell.

--iw 1: The Default Compromise

The subject has aged noticeably — now appearing mid-40s, with visible facial lines and a more complicated expression. The hair has grown out and become disheveled. The rich fabrics remain, but there's tension now between the luxurious styling and Francis' face.

This is where the two forces genuinely wrestle. The output is interesting precisely because it's unstable.

--iw 1.25: The Tipping Point

The brocade coat is now fraying. The ruffled shirt looks worn. The subject is gaunt, the face more deeply lined, the hair longer and wilder. The "fancy" in the prompt is being actively ignored by Francis D's working-class energy. The skin tone has lightened considerably.

I think this is arguably the most cinematically interesting value in the range.

--iw 1.75: Image Reference Wins the R/Face

The reference subject's bone structure and expression have substantially transferred. The fancy costume is now a dark checked coat over a paisley shirt: still "dressed up" by the subject's own standards. The text prompt has been reinterpreted through Francis's world, not the other way around.

--iw 2.25: Clothes Style Takes Over

The subject is now clearly similar to Francis D. The "fancy photoshoot" has been fully recontextualized: he's now in a brown corduroy blazer and dark scarf, sitting in a posed studio setup. The model has found the most plausible version of "fancy" that could contain this specific human being.

--iw 2.75: Near-Perfect Likeness

Strikingly close to the image reference. Same angular face, same hollowed cheeks, same intensity. The background has returned to near-black, mirroring the source. The text prompt now functions almost like a mild style instruction layered over a near-clone.

--iw 3: Image Reference Dominates

The output is basicllay the image reference with minimal creative transformation. Same face, same coloring, same general energy. The "fancy photoshoot" prompt has reduced to a compositional instruction: more intentional pose, same man. MJ has re-rendered the source with modest enhancement.

Sweet Spots & Danger Zones

My opinion: sweet spots are between --iw 1 to --iw 1.5. This is where creative tension produces the most unexpected and compelling results. Neither prompt nor image fully wins, and the results are often more interesting than either source alone.

For likeness-driven work, --iw 2.25 to --iw 2.75 are the best in character consistency, subject matching, portrait series. This range gives you strong identity transfer while retaining enough prompt influence to guide style and mood.

Danger zone: --iw 0 to --iw 0.25 You're uploading a reference for essentially no reason. Save the context, drop the image, just write the prompt.

Files

Comments

No comments found for this post.