How else can AI video evolve? Luma AI’s answer is somewhat unique. While competitors are perfecting individual techniques, they’re like Feng Qingyang teaching the Nine Swords of Dugu – emphasizing flexibility and fluidity, going wherever they please.
Though Luma AI might not be as well-known as Pika or Runway, it’s definitely in the first tier in terms of capabilities. Their video model, called Dream Machine, launched in June this year and recently received a major upgrade – the biggest since its release.
The update includes two major features: first, the release of their own image model Luma Photon, which handles text, images, and video all in one; second, the creation of a new AI video generation workflow where we can generate videos through conversation like ChatGPT, without meticulously crafting prompts.
For video generation, Dream Machine has pioneered a completely new approach.
Link👉https://dream-machine.lumalabs.ai/
Creating Videos Through Conversation
Opening Dream Machine, we first see the “Board” – think of it as an infinite creative canvas where we can freely generate images or videos using natural language.
Starting a new Board, I wrote a very simple prompt: “Create a character from a Japanese mystery manga.”
The AI expanded my prompt and generated 4 images, but the era wasn’t right.
No problem – I just typed another line in the chat box, “Put it in a modern setting.”
The AI understood and generated 4 more images. The top right one, showing a lone figure in a dark alley, was closer to what I wanted, but still not quite there – I wanted him to lift his head and show his features.
Easy enough – just keep fine-tuning, forgetting complex prompts and being direct with the AI.
[Content continues with detailed examples of the iterative creation process…]
Just create – as Luma AI says: “No need to write complex prompts, just ask.” The future questions AI leaves us won’t be about technology, but about what we build with it.
More Controllable Images, More Controllable Videos
Dream Machine’s new workflow is like a tree trunk, with its fun and practical features like branches – complementing each other perfectly to create a flourishing system.
The key component is Dream Machine’s newly released image model, Luma Photon.
How can we generate images that better match our aesthetic? Dream Machine supports style reference and character reference features.
For style references, we can import our own images, and the AI will incorporate that style into the creation. The official example shows generating a bird version of “Girl with a Pearl Earring” in Mondrian style.
[Content continues with detailed examples of style and character references…]
Visual Thinking, Storytelling Methods
Coincidentally, Runway also just announced their image generation model Frame, on the same evening as Luma’s update. The demos show a strong focus on aesthetics, and it’s gradually opening access to Gen-3 Alpha users.
▲ Runway Frame
Image generation quality, aesthetic sophistication, and visual style consistency and control are increasingly emphasized by visual models.
This is good news for creators. When we use AI for graphic design and character development, we’re essentially generating a unique world and telling a unique story.
Runway CEO Cristóbal Valenzuela believes Runway isn’t an AI company but a media and entertainment company, saying the era of AI companies is over.
He’s not downplaying AI – quite the opposite. He believes AI is infrastructure, and the real revolution isn’t in the technology itself but in what it enables: new forms of expression, new ways of storytelling, new methods of connecting human experiences. This aligns perfectly with Luma’s evolution.
[Article continues with discussion of visual thinking and future implications…]
The bugs in Luma AI are still numerous, including but not limited to: character consistency using a single image reference isn’t ideal; credits disappear quickly as you need multiple attempts for images and videos; while the image model can generate accurate English text, Chinese doesn’t work well…
But the message is clear – focus less on parameters and build an AI creative tool centered on interaction.
A better video model isn’t just about faster generation speed, more controllable camera movements, or unique aesthetics – it should also provide better ways of storytelling, using text, images, and video as tools for expressing ideas.
Dream Machine – living up to its name.