AI video is strong now.

You can type a prompt and get a dragon, a city, a fight scene, a product ad, a music-video style clip.

Then you try to direct it.

The camera floats somewhere random.

The character changes between shots.

The motion looks cool, but it does not feel planned.

The scene looks expensive for three seconds, then you realize the model directed the shot for you.

That is the problem with prompt-only AI filmmaking.

The current meta is moving toward reference-based filmmaking.

You collect the visual language first. You make a clean start frame. You block the camera in Blender. Then you feed the image, camera reference, and prompt into a video model like Seedance 2.0 or Gemini Omni.

This is the bedroom version.

I am Alex, welcome to ShortCu8 by Innov8.

Lets Dive Deep 🐰

Use this workflow:

The key idea:

Don't prompt a video.

Package a shot.

Start with FrameThrower.

FrameThrower is a cinematography search engine for movie stills. It lets you browse or search frames using things like genre, shot, angle, time, light, lens, rating, and year.

That matters because film language is hard to invent from memory.

If your scene is "a lonely man walking into a neon street," don't open the video model first.

Search references first. Look at the frame:

Use movie references for shot language, not for copying an exact frame.

The value is in learning how the scene is built.

After references, make the image that starts your shot.

This is where image models matter.

Midjourney V8.2 is best for taste.

Also u can use GPT image 2 // Seedream 5.0 Pro (just released)

The start frame should be clean build it based on the references from framethrower you selected:

Prompt your scenes using any models ( according to the storyline u want)

then create the reference images.

Do not make the start frame too busy.

The video model will add motion, atmosphere, and mistakes.

This is the part most beginners skip.

You do not need advanced Blender.

You need a rough blockout.

Make simple shapes: floor, wall, tunnel, castle, road, character position, and camera path.

Then animate the camera movement.

The Blender render can look ugly. Grey boxes are fine if the camera movement is clear.

The job is camera control.

A text prompt can say:

"continuous forward tracking shot over a castle"

But a rough Blender video can show the model the actual camera path.

That is why Reid Hannaford's Seedance examples are interesting.

The Blender reference is simple, but the final video has a stronger shot because the model has a motion guide.

They still need help with exact direction.

Blender gives direction.

The video prompt should give the model jobs.

Use this structure:

Seedance 2.0 is built for this kind of reference stack. It supports text, image, audio, and video inputs. 4 to 15 second clips and can use multimodal references, including video, image, and audio clips.

Also Seedance 2.5 might get released this week..

Gemini Omni is also moving in this direction. You can also use this model

For one AI film shot, prepare this:

That is the current sauce: less random prompting, more directing.

Now go and create something great

We read every reply - just reply to this email and let us know how we can improve !

#AD1

The AI tool you tried last quarter waited for a prompt, hallucinated a number, then asked if you'd like a summary.

Viktor opened a PR at 2am, rebased it against main, ran your test suite, and posted a note in #eng: "Two flaky tests in payments service, both pre-existing. Recommended merging after fixing them." Then drafted the customer reply for the support ticket the bug created.

That's 619K autonomous actions per day across 20,000+ teams. Not chat replies. Real work shipped to GitHub, Stripe, Linear, Notion, and 3,000+ other tools, from inside Slack and Microsoft Teams.

You don't supervise him any more than you supervise a senior engineer.

SOC 2 certified. Your data never trains models.

"It's what you probably originally thought AI was going to be when you first heard of it in sci-fi movies." Tyler, CEO.

Get Started for Free

#AD2

Your team is already talking about the launch you missed. TLDR is the 5-minute daily brief that keeps you ahead, curated by ex-Google and Anthropic engineers. Free, and read by 7M+ subscribers.

Subscribe for Free