KAIROS

What if a machine could show you the consequences of your decisions before you make them? Not predictions or probabilities, but actual glimpses into the futures that branch from every choice you face today.

KAIROS is that machine. A sculptural device of brushed brass and smoked glass that, once activated, projects holographic timelines for each possible path. You see yourself boarding the train and what happens next. You see yourself walking away and where that leads. You watch yourself answering the phone or ignoring it, and experience both outcomes before committing to either.

I wish KAIROS existed because decisions are irreversible, but our understanding of their consequences shouldn't have to be. To bring this concept to life, I created a cinematic video that demonstrates KAIROS in action: a protagonist faces real decisions, the story splits into parallel paths, and the viewer experiences multiple outcomes, exactly as the machine would show them.

Process Steps

1. Concept & Keyframe Generation Started by generating the KAIROS device using GPT Image 2, defining its aesthetic as a sculptural, premium object. Then created keyframe images for every critical scene: protagonist at a rain-soaked train station, phone close-ups, city streets at night, hotel interiors. GPT Image 2 was also used in image-to-image mode to define the initial and final states of each video clip, ensuring visual consistency between frames and smooth transitions across scenes.

2. Image-to-Video Pipeline Connected keyframe images to Kling 3.0 Omni video nodes. One source image often feeds multiple video branches, the same train station frame generates both the "board the train" and "walk away" clips, maintaining visual coherence across decision paths without redundant generations.

3. Audio & Sound Design Generated a full dark cinematic ambient soundtrack with ElevenLabs Music. Created purpose-built SFX nodes: holographic UI appearance, button press clicks, VHS rewind distortion, electromagnetic activation, and cinematic reveal tones.

4. Voice Cloning for Consistency Used audio-to-audio voice change nodes to maintain the same character voices across every narrative branch. No matter which path the viewer explores, characters sound identical, critical for a branching story where scenes were generated independently.

5. Stitch Nodes for Path Assembly Combined generated clips into complete sequences for each narrative branch using stitch nodes, train path, street path, rewind sequence, and end card, keeping the full assembly pipeline within Melius.

6. Agent-Assisted Workflow The Melius agent was a core part of my creative process. It responded fluently in my native language (Spanish), which made the entire experience feel natural. I could ask questions, discuss creative direction, and get technical guidance without switching mental context. Beyond answering questions, the agent actively did work for me: creating nodes, configuring parameters, and suggesting model pairings. One clear example, I couldn't find a straightforward way to download all my generated assets at once from the canvas, and the agent handled the entire download process for me, navigating the steps I would have struggled to find on my own. It felt less like using a tool and more like collaborating with a creative technical partner.

Feedback

My overall experience with Melius was genuinely impressive. The entire flow creation process felt easy and intuitive, configuration is very visual, and I could follow the sequence of creation at a glance without getting lost in menus or settings. Having multiple models and configurations available on the same canvas made iteration fast and painless; I could try different approaches for each clip and quickly compare results without losing previous outputs. The way you interact with generated results was especially well designed. I could test various prompts on the same node and keep every version accessible, which removed the fear of experimenting.

As for areas that could improve: I'd suggest an easier way to toggle between the cursor modes (arrow vs. hand), perhaps holding Ctrl to temporarily switch, because changing it through the UI didn't feel as smooth during fast workflows. I wasn't able to find a way to batch-download all assets from my canvas at once (though fortunately the agent solved this for me perfectly). I'd also love to see nodes that allow replacing audio tracks within a video (useful for voice cloning integration), merging audio onto existing video clips, or even a lightweight video-editor node that lets you assemble individually generated assets with more control, the stitch node is a great start but felt limited for complex multi-clip sequences.

But overall, my experience was incredible. Thank you for building a product like this, it genuinely enabled something I couldn't have made otherwise. https://app.melius.com/projects/e687aee7-7468-4903-9391-f99a30aa22f8/canvas/0b13fbc1-7eca-4d8a-bb95-98f714176125