Writing
Each monologue is crafted by a person. This stage of the process is about one writer, one voice. No AI impurities go into the scriptwriting process.
Welcome to Doodle Drum
This is the AI-first production studio behind The Improper Doodle. The heart of the series is its writing, which is completely human-created. The grievances, observations, and punchlines are real. This series is the result of extensive artistic exploration and technical experimentation. Learn more here.
AI filmmaking is in a transitional period. While we respect everyone's opinion about the use of AI in art, we believe that it will have as big of an impact on film as postmodernism did. And by the way, in scale and approach, this series is more reminiscent of a comic strip than a film. Read more in the FAQ.
01 / Development
Series Development
Here's a selection of videos to illustrate the evolution of The Improper Doodle.
Summer 2025
Nina is a real Doodle from Kent, England. She lived most of her life in Chelsea, London. Now, she is living in Chelsea, New York. The series was born out of the culture shock she experienced when she realized Manhattan was a much filthier place to live.
Fall 2025
Nina was the perfect test subject for experiments in AI video generation. This early video is cropped to remove the watermark and it struggles to retain consistency across shots. The lip-syncing is rough too. This video was generated using Veo 3 from Google, upscaled with Topaz, and the voice was generated with ElevenLabs.
Fall 2025
Even as AI video has improved, the models struggle to create a female dog, as you’ll notice in the last shot. You’ll also notice that the monologue is rushed because the clips are too short. This experiment felt like the peak of the early experiments.
Summer 2026
We ran a bunch of tests when Seedance 2.0 launched. The model was strong, but the generations were too expensive for the iterative, writing-centric process the series demanded. When Gemini Omni dropped, it was clear that the cost per generation was reasonable. It was time to produce a slate of episodes.
02 / Production
Production Overview
Storyboarding
Arranging shot sequences presents challenges. It's important to run text analysis, choose camera angles, and make cuts to serve the story.
Editing
Finding the rhythm, pacing, and comedic timing is yet another process done by a person. It leads to countless rewrites and regenerations.
Music
Each monologue is set to the same mischievous loop. It's a royalty-free beat created by a human, and lightly edited by another human.
Made with AI
Images
As of July 2026, all the images are generated using Nano Banana 2 and GPT Image 2. Obviously, this will evolve over time.
Video
As of July 2026, all videos are generated using the Gemini Omni Flash model. Of course, this will change again and again.
Voice
As of July 2026, The Improper Doodle’s voice is generated using Eleven Multilingual v2. This is far less likely to change in the near future.
03 / Filmmaker
About The Filmmaker
Theo Miller
Theo Miller is a creative professional who tells stories with emerging technology. As a Senior Designer at Google, he understands the limitations and possibilities of artificial intelligence firsthand by designing software powered by frontier models. Before joining Google, he produced branded content for large tech companies like Lyft, Klayvio, and Carta. Theo's excited to integrate these experiences with films that are personally meaningful to him.
LinkedIn04 / Watch
The Improper Doodle
A British Goldendoodle examines American culture one grievance at a time. Follow the series on Instagram Reels, TikTok, and YouTube Shorts.
05 / Questions
FAQ
- I’m interested in sponsoring an episode of The Improper Doodle. Who do I contact?
- We'd love to hear from you. Email us at theimproperdoodle[at]gmail.com.
- What’s your stance on the use of AI in the arts?
- We believe that AI will lead to a renaissance in auteur filmmaking. Furthermore, this series would not exist without AI. It would be impossible to achieve this level of production value for niche content like this. Funding does not exist for 3D projects about neurotic British Doodles. While we understand there are concerns about data sourcing, these models make this series possible, and equitable licensing models are emerging.
- What models do you use?
- As of July 2026, the images are generated with Nano Banana 2 and GPT Image 2, the video with Gemini Omni Flash, and Nina’s voice with Eleven Multilingual v2 from ElevenLabs. Early experiments in the series used Google’s Veo 3. All upscaling is done with Topaz. Rest assured that the stack will keep evolving, but the writing and music will remain human.