SceneFlow

Turn your script into stunning video scenes

Comment

Inspiration

I’ve always loved the idea of turning a single thought into cinematic storytelling. But creating video content from scratch is hard — especially when it involves writing aesthetic prompts for AI tools like Adobe Firefly.

Most people don’t know how to phrase a good visual prompt, and that creates a creative block. I wanted to eliminate that barrier. That’s what inspired SceneFlow — a tool that takes one idea and turns it into 3–5 ready-to-generate scenes with no extra effort.

I wanted creators to start with a feeling, not a prompt, and still get beautiful B-roll visuals out of it.

What It Does

SceneFlow is a creative Express Add-on that:

  • Takes a simple user idea (e.g., “a sunset walk by the beach”).
  • Breaks it down into 3–5 cinematic scenes using the Gemini API.
  • Auto-generates aesthetic prompts tailored for Firefly or Pexels.
  • Returns clip thumbnails and descriptions in a timeline format.

Here’s the cool part: the user doesn’t need to write any prompts. I trained Gemini to understand each scene using structured prompt templates, so one click gives a coherent, cinematic result.

Users can preview the scenes and use them in Adobe Express designs — without ever leaving the platform.

How I Built It

  • I used the Gemini API to convert abstract ideas into structured scene descriptions.

  • Each scene is built using predefined prompt scaffolds, like:

  Shot Style: Wide cinematic | Mood: nostalgic | Scene: {description}

  • If Firefly support is unavailable, I fall back on the Pexels API for matching stock clips.

  • I built the UI using the Adobe Express Add-on SDK, with React and Spectrum components.

This architecture supports:

  • Text input from user
  • Backend call to Gemini
  • Scene outputs with download links

Challenges I Ran Into

  • No Firefly API access for video made direct generation difficult, so I simulated clip outputs using fallback sources (Pexels).
  • The Express Add-on SDK doesn’t support video injection into timeline, so I designed the output around downloadable or embeddable assets.
  • Prompt shaping was tricky Gemini is flexible, but I had to guide it to output scene-level breakdowns instead of just descriptions.

To make it more reliable, I fine-tuned how prompts are constructed using a form of template chaining

Accomplishments I'm Proud Of

  • I went from idea to working prototype within days complete with full backend and frontend integration.
  • I successfully removed prompt-writing from the user workflow.
  • I proved that AI can act as a scene director, not just a tool.

What I Learned

  • How to build within the Adobe Express Add-on framework, and navigate its limitations.
  • How to translate narrative storytelling into generative prompts effectively.
  • That fine prompt scaffolding makes a huge difference in visual output consistency.

What's Next for SceneFlow

  • Support editing and reordering scenes inside the panel.
  • Extend support to 15s and 30s reel formats.
  • If Firefly APIs are released, integrate direct generation of 5s video clips.

SceneFlow will continue to grow into a full “idea to short video” tool — all within Adobe Express. I want creators to storyboard reels or B-rolls instantly, without technical skill or prompt writing.

Share this project:

Updates