RUNWAY GEN-3 PROMPT STRUCTURES + Midjourney Podcast Launch

Fam,

So last week went nuts…and this week was slower.

But wtf else do we expect in the world of AI.

Lots going on - So let’s not waste time.

Here’s what we got for today:

1. RUNWAY GEN-3 RELEASED

2. RUNWAY GEN-3 PROMPTING TIPS + Structure

3. REIMAGINING AD CREATIVE SYSTEMS WITH AI

4. PODCAST LAUNCH - MIDJOURNEY: FAST HOURS

5. BEYOND MIDJOURNEY: THE DEATH OF UGC/INFLUENCERS?

6. MIDJOURNEY OFFICE HOURS

Let’s Boogie.

RUNWAY GEN-3 RELEASED (TO EVERYONE)

So Runway didn’t “SORA” us.

»GEN-3 is now AVAILABLE FOR EVERYONE → Try it HERE«

As promised…they released the GEN-3 about (7) days after announcing it.

IMO - This is the best text-to-video model available.

But there’s a few things we have to talk about.

If you missed it…FULL LINKEDIN POST

GEN-3 THOUGHTS:

It has range…but it definitely has some rough edges.

I expect this to improve significantly (in short order).

GEN-3 PROS:

• 10-second video gens

• Natural motion is top-of-class

• Prompt coherence is strong

• Advanced camera motion is wild

• Gen time is surprisingly fast (30-45 seconds)

• Very creative outputs

• Handles multiple subjects well

• Human faces + skin texture are very detailed

GEN-3 CONS:

• Text-to-video only (for now)

• No image referencing yet

• Very expensive ($1 for 10 seconds)

• Very reliant on good prompts (also a PRO)

• No “Director Mode” control yet.

WHAT I’M LOOKING FOR NEXT:

IMG Reference → Would expect this release to come in some capacity.

Director Mode → More advanced shot/motion control is possibly next.

Again, they’ve always focused on giving the user CONTROL.

So be on the lookout.

RUNWAY GEN-3 PROMPTING TIPS + STRUCTURE:

If you can text prompt Midjourney…you can prompt Runway.

With that being said…you need to consider 3-dimensional space.

(Not a 2D flat image)

So camera movement is AS IMPORTANT as the subject’s action.

Let’s take a look:

PROMPT STRUCTURE:

Just like MJ…

• Clear + Direct Prompt = Clear + Direct Output

• Ambiguous Prompt = Ambiguous Output

The easiest way to get consistent results is with a prompt formula.

This will give you structure → But also freedom to create.

SIMPLE CATEGORIES:

• SHOT TYPE: How it’s shot (ex. closeup, medium, drone)

• CAMERA MOVEMENT: Camera motion (ex.zoom, panning, orbiting)

• SUBJECT + ACTION: Who and what they’re doing (ex. man running)

• STYLE: Blanket style for the output (ex. Cinematic, anime, video game)

Beginner Prompt Structure:

[Shot Type/Camera Movement], [Subject + Action], [Style]

Prompt Example: 

{Zooming out, zooming in, 360 degree orbiting, panning} of a man on a mountain top, cinematic style

This structure is stupidly simple…but you’ll get consistent outputs.

GOING FURTHER:

The same way we break down MJ images into components like (shot type, composition, colors, lighting, etc).

It’s the same with Runway…use the same thought process.

What elements can you control…and how do you structure them?

INTERMEDIATE CATEGORIES:

• SHOT TYPE: How it’s shot (ex. closeup, medium, drone, tracking)

• CAMERA MOVEMENT: Camera motion (ex.zoom, panning, orbiting)

• CAMERA SPEED/MOTION: Shot pace/movement (ex. hyperspeed, slow motion)

• SUBJECT + ACTION: Who and what they’re doing (ex. man running)

• SETTING: Scene details + emotion (ex. mountain top, moody cafe)

• LIGHTING: Light description (ex. time of day, quality, direction)

• STYLE: Blanket style for the output (ex. Cinematic, anime, video game)

Intermediate Prompt Structure:

[shot type], [camera movement], [subject + action], [environment], [lighting], [style]

Intermediate Prompt Example:

aerial view, camera tracking 2 horses running on a white sand beach next to tranquil water, twilight lighting, 35mm style

HAVE SOME FUN:

These are basic guidelines → Break the rules get creative.

The platform has limitations, but you can test some different items.

Here’s a random prompt structure to play with…

Random Prompt Structure: Starting from a ground-level view of an [environment] leading towards a tunnel inside a [object], the camera tracks smoothly along the [subject] into a short dark tunnel. As it emerges on the other side, the camera rapidly ascends, revealing [new environment + details]

Random Prompt Example: Starting from a ground-level view of a desert road leading towards a tunnel inside a mountain, the camera tracks smoothly along the road into a short dark tunnel. As it emerges on the other side, the camera rapidly ascends, revealing the road continuing to a beach dirt road emerald colored water on the left and palm trees on the right

There’s a lot to test here and it’s currently very expensive…

But I will do it.

Just let me know…

REIMAGINING THE AD CREATIVE PROCESS WITH AI

12 months ago…I had a conversation with SUPERSIDE.

The ideas were BIG → The execution needed to be BIGGER.

We wanted to reimagine the ad creative process for Media Buyers.

Focusing on creating high-end ad assets…in large quantities.

Here’s a little behind-the-scenes of what we created…

FULL POST

WHY THO?

I come from the e-comm world and I work with A LOT of media buyers.

They all have the SAME PROBLEM → LIMITED CREATIVE.

Sooo…why do Media Buyers need options?

→ More options = Better testing
→ Better testing = More optimization
→ More optimization = Better returns
→ Better returns = happy clients/stakeholders

VOLUME = Happy Clients/Stakeholders

So here’s how we solved a volume problem:

FAST FORWARD TO TODAY:

This process set the baseline to create SOP’s for multiple teams.

We took this thought process and applied it to…

Ad/Email Creative, Custom Model Training, Branding Services, Video Production, 3D Design, Motion Design, Web Design, to name a few.

Resulting in…

• 200% faster completion from brief-to-delivery

• 40-85% reduction in hours spent.

• Saving clients over $1.4 Million.

• For companies like Amazon, Reddit, and Salesforce.

If you’re looking for fresh and efficient…high-quality…DONE-FOR-YOU design + creative…hit the link below.

PODCAST LAUNCH - MIDJOURNEY FAST HOURS

So Drew Brucker and I started a podcast…about Midjourney.

We have NO IDEA what we’re doing. (Cut us some slack)

Here’s the first look at MIDJOURNEY: FAST HOURS

WHY THO?

Every AI podcast is the same…

It’s either “Breaking News” or “Here’s HOW TO (X)”

Feels like it’s lacking substance and the human element…so we’re going to have some fun with it..and create a space for the MJ degenerates to nerd out.

How we’re thinking about it:

• Less structure…more conversation.

• Less news…more practical application.

• Less formality…more entertainment.

THE PLAN:

We’re going to experiment with it → Try different formats, segments, and topics.

Open-source it (in a sense)…All feedback/ideas/criticism are welcomed. 

BEYOND MJ: LIVEPORTRAIT + THE DEATH OF UGC?

This is where things get weird.

Is this the death of UGC + Influencers? (It might be)

LIVEPORTRAIT is an open-sourced model that allows you to animate still images by using a source video for realistic facial movements.

GITHUB LINK: https://github.com/KwaiVGI/LivePortrait

NEW FRONTIER

As always…this is cool…but it’s concerning (2 things can be true)

The lines between fiction and reality will be blurred…and this has “deepfakes” written all over it.

If you need help conceptualizing it further…I’m pretty sure James Cameron spent millions on this tech…but now my stupid ass can do it from my couch.

(That’s a gross exaggeration…but you know what I mean)

MIDJOURNEY OFFICE HOURS (7/3):

Website Updates

• Personalization: New onboarding feature, syncing daily theme rooms, room reorganization.

• Launch: Nearing full release despite imperfections.

Model Releases

• Next 6.5-like release split into 2 releases

  • First release focused on pixel quality

  • Second release targeting hand coherence, body coherence, & text accuracy

  • Versions might be called 6.1 & 6.2

Personalization

• Working on personalization versioning system

• Developing personalization-powered explore page

Have fun.

If you need help…DM me on Linkedin.

Until next time…

Rory