❔ About
With this post you'll see how I started my first full artwork creating a bridge between:
- 📜 Data
- 🎙️ Sound design
- 🤖 Generative Text To Speech
- 🖌️ Video artwork
- 📢 Digital contents streamline
- 📈 Social networks and content embedding
💡 Inception
What triggered this creation is the following tweet:
... I immediately started to think:
"... and if I could create a fully digital, multimodal Customer Experience that would be ready to be shared on social platforms ?"
☝️ Also, people are talking a lot about about AGI like Midjourney, DALL-E... but very much less about Generative AI for TTS (Text to Speech).
♾️ " Voice prompts," aka. "History prompts"
As all others AGI, suno-ai/bark makes no exception : it relies on "PROMPTs".
Luckily, the bark's community is very active and share their voices prompt (and tags) discoveries :
🔁 Creative workflow
Here is the current workflow I could experiment:
- Create & release a SDK to get the data
- Imagine a customer experience at restaurant
- Develop & tune the data driven script and build soundtrack
- Create an avatar and scene for the waitress
- Put together soundtrack & avatar into video
🧰 Tools
Here are the open source tools I used for now:
-
📦
auptitcafepackage to get the data -
🎨
bluewillow.ai: "AI Artwork Generator" for avatar creation -
🐕
suno-ai/barkto build effective soundtrack -
🤏
pydubto compresswav(~ 20 Mo) tomp3(825.65 kB) &webm(1.68 MB) - 🎥
OpenTalker/SadTalker - 🦾
T4x2 GPUfrom Kaggle
🍿 Demos
Below are the demos:
🤓 How it's built (author's words)
🎙️ Soundtrack
Output soundtrack with bark:
🎞️ Movie
Then put the sound into an avatar with SadTalker:
🤔 Ideas for "later"
Automate:
- Video creation
- Video upload on dedicated cloud services for further optimal collaboration, digital marketing,...
- Avatar creation so video is totally code driven... and makes content more original (and funny) on each release thanks to one time generative prompt (prompt design required)
↩️ Conclusion
The more I think about designing - and achieving - such experiences, the more I find evident the core of this kind of project is:
- 🎯 Get a clear idea and be strongly focused on what you want to achieve (ie. you don't get lost in your creative journey)
- 🔗 Design a clean linear workflow that focus on tasks (not tools) so you can adapt it easily as AI projects are evolving at an amazing pace (I mean every week there are new tools)
🔖 Resources
🔭 Tools to prototype
Top comments (9)
Announcement on Github discussion