Creating Cinematic Images & Films with Generative AI
London
- Tweets 3,152
- Following 777
- Followers 2,141
- Likes 13,825
Pinned Tweet
Here’s Part 2, this time made with Seedance 2.5.
This is the final cut, built from shorter generations, and I’m really happy with how it came together.
My first attempt was one continuous 30-second generation, but it introduced too many mistakes. At 2.5’s price, repeatedly rerolling a long clip just wasn’t practical.
Breaking the sequence into shorter shots gave me much more control over blocking, character placement, vehicles, pacing and the final edit.
The biggest upgrade is the rendering. Skin has more texture and looks far less plastic than it often did in 2.0.
Multimodal referencing is genuinely useful too. I used character, vehicle and environment sheets, but never needed more than seven references at once.
Is it worth the extra cost? Yes, if time matters. I think 2.0 could reach a similar result, but it would take more generations and a lot more work.
A few example prompts in next posts 👇
Made via @dreamina_ai
Second go with @Hailuo_AI #MiniMaxH3
This time I ran gritty Midjourney v8.2 stills through H3 and used a Suno track as reference.
Big takeaway: simpler prompts worked better. H3 understood the intent, stayed close to the original aesthetic, and synced the movement to the track surprisingly well.
There are a few mangled faces and hands in there, but most of that came from imperfections in the source stills rather than H3 breaking the motion. This was just a quick test, so I left them in.
I also used a vérité-style camera block and crash zooms to make the footage feel less controlled and more observed. Handheld movement, slight focus breathing and subtle background drift gave the shots a much more natural feel.
Prompt block for the vérité camera work in the next post 👇
49
11
257
49,312
Happy to share that I’ve joined the CapCut Creative Partner Program.
Grateful to the @capcutapp team for the invitation and the support for my work.
Looking forward to seeing what I can make with their support. #CapCutCPP
14
71
1,857
V retweeted
A short story "Sail Anyway". A reflection of the thoughts that've been on my mind lately. Let me know what you think.
42
28
504
95,517
V retweeted
Walter
My submission to the Lumara Film Festival.. special thanks to @AskVenice
16
7
46
6,777
V retweeted
Two Fwnds
"Nobody ever says the right thing."
Everyone knows this. Most of us learned it the hard way when we're on the receiving end. But we still can't help ourselves, because the hardest advice to take is the type we're best at giving.
"I've told two people tonight."
An hour before a Midnight Snack. Second in a series.
Everyone thinks the considerate thing to do is ask when someone seems to be hurting.
Not always.
Asking is for you.
You get to feel like you helped,
and the discomfort moves off you and onto them.
So he does the one thing she requested: nothing.
#dreaminacpp @dreamina_ai
8
3
35
4,713
V retweeted
“The Miracle” Premieres Now:
I gave this film everything I had. Pushed myself creatively and technically, experimented a ton with new tools + models and at times got stuck, started over and wondered if I could pull it off. Excited + nervous to finally share it with you. Enjoy!
123
86
576
116,883
V retweeted
@LumaLabsAI Uni 1.1 is doing something with camera prompts that I wasn’t expecting.
Super 16mm + Angenieux 11.5-138mm HR T2.3.
The prompt adherence is incredibly strong, and more importantly, it doesn’t just add “film grain.” The whole image starts to feel like it was actually shot through a lens.
For almost all my recent work, this is the camera lens I PROMPT in...
Often, it's all I prompt for aesthetic direction, meaning no descriptions for how it looks... and it nails it. Try this with the @LumaLabsAI Uni 1.1 model.
Add "1990's camera aesthetics:" or any decade to the beginning of this prompt for a more specific look.
CAMERA:
Super 16mm film and an Angenieux 11.5-138mm HR T2.3 S16 PL Mount S/N 1566465 lens.
4
5
67
3,670
V retweeted
Two weeks of intense work! 🔥
Six minutes, as intense as I could make them!
Fellow creators, I poured my very soul into this.
If you look closely, you will see me there.
And maybe you'll see yourself in it too.
INCUBUS. A story I carried in my head for eleven years: a man who can leave his body and live inside other people's. Unfilmable, until now.
This film exists because of the festival Higgsfield launched! I love the energy it has unleashed, all these talented creators I discover here every day! ❤️ I didn't make this to prove what AI can do. I made it to reach you. Let's make work so strong that nobody cares whether it's AI or not. Let's make the world dream, together.
Watch it. Then tell me what it did to you. And if you like it, support it — comment, repost. I have even more ambitious films in me, and I fully intend to spoil you. Thank you, my friends! 🫶 #HiggsfieldGFF
@higgsfield_ai
195
234
1,606
100,287
....soon🍿
Did you see my film The Miracle in todays Paper? Some of my favorite creatives got an early peak at the film + the reviews are coming in Hot!
In case you don't still get the newspaper at your place here is the ad from todays paper. Film premieres next week!
@CaptainHaHaa @DrDreams @BLVCKLIGHTai @tupacabra @stinkycubert @JoshDaws @jordandchesney @VictorInFocus @mike_darrow @bstuartTI @eattheethos @jdp2oo
21
720
Love this from @xdeconstructed
The characters seem familiar and natural, like people you might actually know. Nate manages to get these really natural, realistic performances out of his characters 👌
Everyone thinks the considerate thing to do is ask when someone seems to be hurting.
Not always.
Asking is for you.
You get to feel like you helped,
and the discomfort moves off you and onto them.
So he does the one thing she requested: nothing.
#dreaminacpp @dreamina_ai
3
25
1,212
V retweeted
Everyone thinks the considerate thing to do is ask when someone seems to be hurting.
Not always.
Asking is for you.
You get to feel like you helped,
and the discomfort moves off you and onto them.
So he does the one thing she requested: nothing.
#dreaminacpp @dreamina_ai
38
22
414
92,432
...taking notes ✍️
PROMPT SHARE!
This is for the 2nd shot in 'Beautiful', where she cries. I used the first shot (the interview) as a single video reference:
Continue @ VIDEO 1:
CUT TO:
A tight telephoto close up on the supermodel's face, capturing her with a voyeuristic telephoto camera shot (passing by people, walking through the frame heavily obscure our view of her) as she sniffles, delicately wiping away her tears. She's been crying. Her eyes are slightly red and face flushed having just finished crying. She tries to compose herself, trying to hide herself from the surrounding crew members walking around (out of focus in the foreground, walking through frame). She is embarrassed to be having such an emotional breakdown in such a public place. White wall of the studio soundstage in the background (Hollywood, California).
The entire shot is one seamless take, captured with an old super 16mm handheld camera with an Angenieux 11.5-138mm HR T2.3 S16 PL Mount S/N 1566465 lens (global shutter natural camera swaying and shaking). The shot maintains a tight telephoto close up for the majority of the shot, lingering on her face.
With a great sense of self consciousness, she looks around at the people walking by her, turning away from them to hide herself so no one sees that she's been crying.
Moments later we hear a man approach her, really concerned for her. Off screen, we hear him say, "Hey... hey, listen I'm sorry. I'm sorry, okay." She immediately turns away from him. She does not want to see him. He is the reason she's been crying. She tries to walk away from him. The tight telephoto camera shot pans, following her (the camera drifting as she walks, having a hard time trying to follow her: the camera never knows where she is going to go, so it's always lagging behind her movements, trying haphazardly to keep her in frame). The man (off screen) follows her and enters the frame, gently taking her by the arm, turning her towards him and stopping her from going. She turns to him reluctantly, completely fed up with him and totally uninterested in what he has to say. He says, "Hey, please." A beat. "Please. I'm sorry, I really am. I didn't mean it, okay, I mean, I didn't mean it to sound like that, I'm an idiot, okay. Please. Please forgive me. Please." His sincerity, softens her towards him. She is still deeply hurt, but doesn't pull away so much, but instead her body language shows that she is becoming receptive to his contrition. Then, as she says, "Why do you do this though?" she can't help but start to cry again, and he immediately holds her close and she weeps, burying her head in his chest.
1
4
542
V retweeted
Follow up from the Wan 3.0 clip.....
At some point Walter White apparently wandered into the cast, went blonde and found a pink suit.
Genuine mistake, by the way. I forgot to add the character sheet and it somehow gave me Bryan Cranston. I never prompted for him.
How did I completely miss @Alibaba_Wan Wan 3.0 dropping?
It released a few days ago and I only just got around to trying it.
MiniMax H3, FLUX 3, LTX 2.5 and now this. We’ve been ridiculously spoiled with video models lately.
Up to 20 multimodal references, 30-sec generations, and just $0.10/s at 720p.
First proper play with Wan 3.0 and I think I love it.
Prompt for this in the next post 👇
11
4
63
5,339
V retweeted
I used ChatGPT to analyse a Fast & Furious chase scene and turn it into better Seedance 2.5 prompts.
The difference was huge.
I was prompting “high-speed chase”, but the shots still felt slow, not dynamic at all.
So I had chat gpt (5.6 Sol on High) study and analyze a reference clip and pull out the actual visual cues of speed:
• start already at full speed
• let the camera get overtaken
• use strong foreground parallax
• let vehicles rapidly change size in frame
• vary the distance between the vehicles
• use short interior inserts as punctuation
• reveal hazards late
I then gave ChatGPT ByteDance’s official Seedance prompting guide and asked it to audit the prompt.
bytedance.larkoffice.com/doc…
Chat GPT revised the prompts around the Bytedance guidance, simplifying the physical action, separating it from the camera direction, and giving each clip one clear action arc.
As you can see from the side-by-side, the camera work shares the same visual language, and it worked well.
To note, my version is several shorter generations edited together, rather than one continuous generation.
The full caht GPT analysis prompt is in next post below 👇
Here’s Part 2, this time made with Seedance 2.5.
This is the final cut, built from shorter generations, and I’m really happy with how it came together.
My first attempt was one continuous 30-second generation, but it introduced too many mistakes. At 2.5’s price, repeatedly rerolling a long clip just wasn’t practical.
Breaking the sequence into shorter shots gave me much more control over blocking, character placement, vehicles, pacing and the final edit.
The biggest upgrade is the rendering. Skin has more texture and looks far less plastic than it often did in 2.0.
Multimodal referencing is genuinely useful too. I used character, vehicle and environment sheets, but never needed more than seven references at once.
Is it worth the extra cost? Yes, if time matters. I think 2.0 could reach a similar result, but it would take more generations and a lot more work.
A few example prompts in next posts 👇
Made via @dreamina_ai
14
2
74
3,701
Love this shirt by @cfryant It’s clean, strange, and beautifully paced!
New short film, "Noise to Signal".
Made on @invideoOfficial's Agent Two.
Over a month in the making, this short film is a showcase piece for my new AI VFX studio, @redacted_owl (Redacted Owl Studio) for film and tv - announcing for the first time, right here and now.
Stay tuned for the "making of" with Agent Two coming soon!
2
15
1,047
V retweeted
Really can't afford to blink these days! 👀
How did I completely miss @Alibaba_Wan Wan 3.0 dropping?
It released a few days ago and I only just got around to trying it.
MiniMax H3, FLUX 3, LTX 2.5 and now this. We’ve been ridiculously spoiled with video models lately.
Up to 20 multimodal references, 30-sec generations, and just $0.10/s at 720p.
First proper play with Wan 3.0 and I think I love it.
Prompt for this in the next post 👇
16
5
101
13,013
