EP 044 · 88 MIN MIDJOURNEY VIDEO DOES WHAT RUNWAY CANNOT
Midjourney Video is Here! Nerds React + New --SRef System Chop-Up
The short version
Midjourney surprise-launches its first video model, and Rory Flynn and Drew Brucker argue it already beats rivals on style: auto-prompt turns any still image into a coherent video with one click, preserving complex Midjourney looks and SREF codes in ways Runway and other generators typically lose. Bugs remain (reversed wheels, cars driving backward, a 480p-only ceiling), and the hosts unpack Midjourney's new --sref system, which now defaults to --sv6 and silently breaks legacy codes unless creators add --sv4.
Key takeaways
- 01
Midjourney's auto-prompt needs one still image, no written prompt, and renders a strong video in under two minutes: Rory called it 'massive' compared to typing paragraphs.
- 02
Complex, highly stylized Midjourney images animate with their look intact: Runway and other generators typically lose that style the moment a shot pans or moves.
- 03
Static wheels and un-signaled direction carry over from stills to video: cars can drift backward or keep non-spinning wheels unless the source image shows motion.
- 04
Midjourney's new --sref codes default to --sv6 and silently break if reused without --sv4: Drew's whole saved code library now needs the old flag tacked on.
- 05
Upscale Midjourney's 480p-only video through Topaz Astra instead of waiting on native HD: Rory helped build Astra and rates it above Topaz's newer Bloom upscaler.
- 06
Midjourney's SREF no longer leaks a reference image's subject into new work: Rory's all-white sneaker test pulled only the jungle photo's color and haze, not its leaves.
- 07
Build SREF test batches from one-word aesthetics (weirdcore, mythpunk) and Spotify song titles, not just 'album cover': Drew's method for finding new styles.
Receipts · said on the record
“When I started hitting the button, I was like, this feels like version 4. … when it hit right away, I was like, thank god. You know, it was just like a wave of relief of like, yes, all right, we're back in it, baby.”
“You can just hit auto prompt and you can have a really cool video in less than, you know, two minutes from a really good aesthetic without having to go and prompt six paragraphs. Right. To me, that was massive.”
“Mid-Journey style would not make it to the rest of the video because it's just too complex, and the video generators are just like, I have no data on this, so I can't compute. Runway notorious for that.”
“So the interesting thing is that if you have messed up hands in an image, the hands correct in the video. I've never had an issue with hands in videos, really.”
“The new SRF system broke the legacy system, meaning if you were to use this code and not attach dash-dash SV space four, you're gonna get something totally different.”
Chapters
- 0:00:00 Thursday night chaos & surprise video drop
- 0:03:40 Midjourney’s whisper video launch strategy
- 0:05:10 A wave of feature updates: SREF, video, pricing
- 0:06:15 Long-term vision: real-time world-building
- 0:07:00 First impressions of Midjourney video
- 0:08:30 Auto-prompt is criminally good
- 0:09:55 High motion vs low motion—how to choose
- 0:12:24 The magical 4-grid video preview
- 0:14:00 Interface brilliance: simplicity done right
- 0:15:20 Observations on motion templates & quirks
- 0:17:00 Animation consistency & style fidelity
- 0:18:30 Old stills, new life—archive resurrection
- 0:20:00 Using SREFs to build entire worlds
- 0:22:50 Where the current video model falls short
- 0:24:20 Enter Topaz Astra—perfect timing for upscaling
- 0:26:00 A personal story on AI, portraits & impact
- 0:28:40 Upscaler comparisons: Magnific vs Bloom
- 0:30:00 Visual bugs: wheels, backwards cars & physics
- 0:32:00 Why hands magically fix themselves in video
- 0:34:00 Face consistency & surprising details
- 0:36:45 Sneaking Drew & Rory into an animated video
- 0:37:30 New Explore page for video (and why it rules)
- 0:38:50 The Extend feature is shockingly seamless
- 0:40:15 Camera direction tips for better results
- 0:42:20 Prompt control vs camera behavior when extending
- 0:44:10 Wild new video prompts spotted on Explore
- 0:47:00 Midjourney’s edge: vibrant color retention
- 0:51:00 SREF update — what broke, what’s better
- 0:54:15 Favorite tokens: dribbble.com, Nike ads, stoicism
- 0:57:40 Using songs, moods, and weirdcore to test SREF
- 0:59:15 How --exp interacts with --sref (and when to dial it down)
- 1:03:00 SV6 vs SV4 — why it matters now
- 1:08:00 Building smart prompt batches for testing
- 1:10:20 Describe tool to reverse-engineer SREF aesthetics
- 1:17:00 Style reference vs subject leakage test
- 1:23:45 Personalization wish list: can ranking images help?
Questions this episode answers
What did Fast Hours think of Midjourney's new video model?
Rory and Drew call it a stronger-than-expected V1: auto-prompt turns a single still into a coherent video in under two minutes, and complex, stylized Midjourney images keep their look through motion in a way Rory says Runway and other generators typically can't.
How does Midjourney's new --sref system work, and what broke?
Midjourney's June 2025 --sref update made --sv6 the new default style-code version; Drew explains that any previously saved --sref code used without appending --sv4 now renders in a different, unrelated style than before.
Does Midjourney's video model have bugs?
Yes: Rory found that cars auto-prompted without explicit direction sometimes drive backward, and wheels that look static in the source image often stay static in the video, though messed-up hand anatomy tends to self-correct.
What is Topaz Astra, and is it worth it for upscaling Midjourney video?
Topaz Astra is Topaz's diffusion-based video upscaler, built on its earlier Starlight model. Rory, who consulted on both, recommends it for Midjourney's 480p-only video output but notes it can over-smooth and over-contrast footage that already looks rough.
How does Drew choose --sref codes to test on the show?
Drew runs short, repeatable test prompts, like 'dribbble.com', 'album cover', 'movie poster', Nike ad creative, stoicism terms, one-word aesthetics such as weirdcore, and Spotify song titles, then upscales and keeps only the strongest handful from each batch.
Show notes
Fast Hours Podcast Ep. 44—Midjourney Video is HERE—Nerds React + New SREF System Chop-Up
Rory’s parents are laughing in the background. Drew just got a Midjourney tattoo that reads "/imagine". And somehow, Midjourney video dropped in the middle of all this.
This week on the Fast Hours podcast, the boys stumble into an accidental emergency episode because (surprise!), Midjourney just casually launched the most fun AI video tool yet.
No press release. No tweetstorm. Just boom, it’s here. And it works. Like, stupidly well.
They run live experiments, break down what’s working (and what’s weird), and unpack the shockingly seamless auto-prompt, motion settings, extend feature, 4-up grid, and how it plays with SREF.
Spoiler: SREF just got way more powerful—and more confusing. Also, Drew finds a way to sneak himself into a talking head video, naturally.
Think you’re burned out on AI? This might be the update that gets you feeling things again.
