Post OS

This is a first try at generating dialogue scenes with Seedance 2.0 from 2 character sheet and multiple location stills.

Findings:
- Limiting location sheets to a single wide shot feels a better strategy to limit random backgrounds
- Voices inside Seedance 2.0 are VERY hit or miss. I still prefer Kling out of the box.
- Negative prompting feels like a must currently on Seedance

Comment 'POST' to start building with AI.

The midwife had delivered every child in the village for forty years.

She never left without saying the child was healthy.

The night he was born she wrapped him, handed him to his mother, and walked out into the winter without her coat.

She walked to the oldest house in the village. Knocked. When the door opened she said three words.

THE Elders knew right away. But were sworn to secrecy.

He grew up like any other man. Fished. Hunted for furs. Ate when there was food left.

But when his eyes went wet — the boats stayed in. He never knew why.

Then one morning something rose from the black water. Something not of this world.

And everything he was born to be finally had a name.

THE RAVEN

Comment 'POST' to start building brands with AI.

The first Seedance 2.0 edit is in, and boy oh boy am I excited for the future of content. This clip is comprised of 4 generations (not cherry picked). I've used a bunch of different methods, form character sheet + location image reference (the best workflow so far), to image + video + prompt, to all out 9 image references plus prompt.

My findings: the model performs better when you leave it the most creative leeway in terms of compositing keyframes and producing multishot clips. Using too many keyframes made each keyframe animate very stiff and we lost a lot of detail. Video to video to simply add back folley and atmospehres is a breeze (sound design for existing clips with no sound is solved at this point). The model breaks colors and detail with video as inputs.

I have yet to test out more action-style scenes, but for slower establishing shots, it's crazy good. The way the camera moves, using cranes and rotation rigs is truly stunning.

Comment 'POST' to start building with AI.

Sneak peak of content from my upcoming movie making course. Stay tuned.

Comment 'POST' to get ahead of the competition.

I really feel a big part of what we call "cinematic" is in fact negative space, or the space surrounding our heroes on screen.

That, and the sense of scale our sets and enemies brings, in relation to said heroes.

When the world feels girnormous and the hero is always well positioned and lit on screen, you get cinematic.

We could get into color theory and design, but this is a MASSIVE one. And so here's the workflow to improve cinematic vertical videos:

- Create your scene in 16:9 or 21:9
- Use Nano Banana Pro for generative fill top and bottom
- Animate, add music, and voilà.

Prompt: remake this exact composition in 9:16 format (simple as that)

Comment 'POST' to build brands with AI.

The art of the sizzle reel is not an easy one. Each shot has to tell an entire story.

Hero > Monster > World > Challenge

Everything is a few seconds. We need to know form the composition who we root for, who we fear, where this is all happening, and what exactly IS happening.

Then there's the sense of scale. When building a killer trailer, you need scale. Massive monsters, massive buildings, massive explosions, and tiny heroes.

I realized there's a version of Midjourney that does a much better job at conveying scale: 6.1.

I never thought I'd go back, but as soon as I tested 6.1, I saw something I hadn't seen in a long while in my generations: fantasy. The incredible world of make-ups, weird skins, monsters.

Comment 'POST' to start building brands with AI.

Here are the workflows to coimplete my latest mini movie "Gong". A total of 44 shots were create and animated, 4 of which served from start/end frames. Can you guess which ones they were?

From 3D title animation, to character consistency, to scene compositing, to dialog shots.

Total production time: 6 hours
Base images: Midjourney
Compositing tool: Nano Banana Pro (inside Weavy)
Animation, sound effects and character voice: Kling 3.0 (outside Weavy $$$)
Soundtrack: Suno v5
Narration: Eleven Labs

Comment 'POST' to start building brands with AI.

The student had heard the stories.

Emperors had tried to defeat him. None returned to speak of it.

He crossed three provinces to stand before this man — if a man he still was. Those who had seen him spoke afterward carefully, the way people speak when they are not sure what they witnessed.

The student had read everything written about him. He had decided, somewhere along the road, that this alone was preparation enough.

The master looked at him once: “Sit. Listen. Return tomorrow.”

The student sat. He listened. He returned.

Three days passed.

On the fourth morning he rose and stood before the master.

“This is not teaching.”

The master did not look up.

The student descended the mountain that same hour. He returned to his village. He married. He tended his father’s farm, and then his father’s father’s farm. The seasons moved through him. He grew old in the way most men grow old — completely.

There was a girl who had arrived at the mountain not long after he had.

She did not leave.

She sat. She listened. She returned.

Through winters that cracked the stone steps. Through years that offered nothing visible. She stopped measuring. She stopped expecting.

Decades later her name traveled differently than names usually travel.

They said she no longer aged. That illness avoided her the way water avoids certain stones. That she had become something the old stories had anticipated but never quite named.

The man heard this on his deathbed. In a small house. On his father’s farm. A single tear ran down his cheek.

Her name was Gong.

Comment ‘POST’ to start building with AI.