EP 039 · 70 MIN OMNI-REF NAILS FACES. WRECKS HANDS.
Midjourney Omni-Reference and --exp: Showing You How They Actually Work
The short version
Episode 39 is a live test lab for two new Midjourney V7 features: the --exp parameter and Omni-Reference. Rory and Drew run --exp across stylize values from 50 to 1000 and find it pushes the subject back from camera and cuts through Midjourney's personalization tint, while Omni-Reference nails facial detail and product texture but mangles hands and can't be combined with editing. They also pit Google Gemini, ChatGPT and Flora against Midjourney for consistent-character and outfit compositing.
Key takeaways
- 01
Running --exp up while stylize is fixed backs the camera away from the subject: the same prompt gets progressively wider as EXP rises from 0 to 100.
- 02
High --exp values (50 to 100) break Midjourney out of a personalization profile's color bias, like the blue-pink tint Drew's global profile adds at high stylize.
- 03
Isolate a face on a plain background before using it as an Omni-Reference: Rory got sharper freckles, nose shape and skin texture than with busy source photos.
- 04
Stack Object Weight 400 with EXP 30-40 when placing a full character into a busy scene: Rory found this combo preserved his face and even improved hand rendering.
- 05
A multi-angle "character sheet" used as one Omni-Reference image locks the output to its framing; dropping Object Weight to about 25-50 lets a new scene's composition take over.
- 06
Omni-Reference blocks editing outright: strip it from the prompt first if you want to touch up a generated image.
- 07
For product-on-person composites, keep Object Weight low (10-25); higher values pull in the reference image's overall aesthetic instead of just the item.
Receipts · said on the record
“I run that up to 250, which I've been running a lot more of, and you kind of start to see where my personalization goes. It's like this blue and pink tint, and I don't like it.”
“I hate to say that we're going back in time, but for whatever reason with OmniRef, the hands are worse. And especially the further away they get from the frame, the more mangled they get.”
“Naturally, his freckles are not going to be one-to-one from image to image. But like, the imperfection on this made it awesome too, because he's got this weird kind of nose, and that sort of shows up in everything.”
“It helped the hands too. Don't ask me why. When I was using it, I was getting better hands.”
“You get a scar or something that maybe is more of a singular thing that feels replicable. You get into things like freckles where it's just more dynamic or there's too much intricacy. That's where AI just in general, I would say, fails.”
Questions this episode answers
What is Midjourney's --exp parameter and what does it do?
Exp is Midjourney's experimental-variation parameter, which Drew and Rory describe as feeling like pre-personalization stylize. Raising it breaks images out of a user's personalization color bias and, at a fixed stylize value, pushes the camera progressively farther from the subject.
What is Midjourney Omni-Reference and how well does it work?
Omni-Reference is Midjourney's V1 replacement for Character Reference, letting users drop in a face or object photo to carry into a new generation. Rory found it excellent at preserving facial detail and product textures but noticeably worse at hands, and it can't be combined with the Editor until it's removed from the prompt.
What Object Weight should I use with Omni-Reference?
It depends on the goal, per Rory's live tests: around the default 100 for simple face-only translations, 400 paired with EXP 30-40 for placing his full likeness into a busy scene, and a low 10-25 for compositing a product like shoes or an outfit onto a person without dragging in the reference's whole aesthetic.
Does a multi-angle "character sheet" image improve Omni-Reference consistency?
Not in Rory's live test: Midjourney locked onto the same framing and distance as the reference collage rather than treating the separate angles as options. Dropping Object Weight to around 25-50 let the new scene's composition take over, at some cost to character fidelity.
Which AI tool is best for consistent-character generation, according to Fast Hours?
Rory rates Google Gemini highest for keeping a subject's identity consistent outside Midjourney. He and Drew also tested ChatGPT and Flora for compositing a generated outfit onto a photo; ChatGPT's result carried an unwanted warm yellow tone while Flora matched the reference colors more closely.
Show notes
Fast Hours Podcast—Ep.39: Midjourney Omni-Reference and --exp: Showing You How They Actually Work (with Drew Brucker & Rory Flynn)
Midjourney just gave us two powerful new tools: --exp (experimental variation) and Omni-Reference. In this episode, we test both — live — and show you exactly how they behave, what they actually affect, and when they’re worth using.
This is your walk-through, deep dive, and test lab all rolled into one.
They also cover upcoming Midjourney features teased during Office Hours — including further aesthetic upgrades, sref and moodboard plans, upscaling, quality rendering upgrades, and more ways to fine-tune your style and control.
