Since September, a diverse group of artists worked with Flow, the AI tool for filmmaking, in a pilot called Flow Sessions. It was two months of experiments, workshops and mentorship where we tested how an AI can be integrated into real creative processes — not just as a tech demo but as a working partner.
1. Adopt the director mindset
What makes a tool powerful? The intention of the person using it. The artists who made the most of Flow thought first about story, characters and art direction, not technical tricks.
Leilanni Todd: The magic happens when you bring your own vision, art direction, storytelling and point of view to guide
Flow— that’s when something truly original emerges.
Technically speaking, that means using the AI as an engine that responds to well-designed inputs: a prompt that includes narrative tone, visual references, shot composition and lighting notes. Instead of asking the AI to generate without direction, the best results came from short iterative cycles: storyboard sketch, test generation, fine-tuning parameters (clip length, continuity, aesthetic) and trimming.
Practical tip: turn your vision into actionable specs. Translate an idea like “nostalgic and warm” into visual references, a color palette, and camera examples (close-up, slow tracking). That improves cinematic and temporal coherence in the generative output.
2. Technical curiosity > expert knowledge
A clear surprise was that technical barriers weren’t the main limitation. Participants had different experience levels, but they all shared one attitude: a willingness to try, fail fast and learn.
Alex Naghavi: The people shaping what’s coming aren’t the ones who know the most, but the ones who dare to experiment.
From a technical point of view, this translates to using interfaces that abstract complexity — presets, sliders for style and timing, and the option to upload references — so creators can try combinations without writing code. For developers, the lesson is clear: UX matters as much as the model. If you want more artists to use your generative tool, invest in workflows that make iteration and quick feedback easy.
Technically useful: think about pipelines that allow a "human-in-the-loop": preview versions, annotations about what failed, and retraining or parameter adjustments without exposing complex models to the user.
3. Tell the untold stories
Flow Sessions allowed many intimate stories to reach a visual form that previously felt out of reach because of cost or technical barriers. Family stories, memories and archival material found new life.
- Chris Carboni transformed recorded conversations with his grandmother into a short film that mixes humor and nostalgia. The original audio was integrated with high-quality generated images to create a “digital legacy.”
- Katie Luo turned real photos of her grandparents in Taiwan into dreamlike landscapes, exploring generational affection and cultural differences.
Here’s a technical and ethical note: when you work with real people’s voices or images, consider consent, usage rights and preserve the original versions. Generative tools make creation easier, but they also amplify risks of decontextualization.
Practical suggestions for creators:
- Always archive source material and record metadata (date, context, permissions).
- Keep a chain of versions so you can revert or audit changes.
- If you use voices or faces, secure explicit and documented consent.
How to apply these lessons if you’re a creator or developer
If you’re an artist:
- Prioritize story and use AI to visually iterate, not to replace creative decisions.
- Start with short clips to validate style and continuity before producing long sequences.
- Use visual references and shot lists to guide generation.
If you’re a tool developer:
- Design interfaces that reduce complexity: presets, reference import, and art-direction controls.
- Enable feedback pipelines that allow adjustments without retraining whole models; think style parameters and post-processing modules.
- Measure not just visual quality, but temporal coherence, inference latency and cost per minute of render.
From the technical side, areas worth investing in are:
- Multimodal models that integrate audio, image and text to maintain narrative coherence.
- Parametric controls for cinematography: shot length, camera movement, lighting.
- Infrastructure that lets you render fast iterations in the cloud with version traceability.
Final reflection
The most interesting thing about Flow Sessions isn’t only what the AI can generate, but how it changes the creator’s role: from tool operator to director of a human-machine collaboration. Do you need to be an AI expert to take part? No. Do you need intention, curiosity and ethics? Yes.
Original source
https://blog.google/technology/google-labs/3-things-learned-flow-sessions