Teller: researching automated content generation from descriptive scripts.

Watch a short video created with Teller (youtube link)    Final animation after animator and composition phase from 3D Artist.  (I’ll try to summarize the project, but this is an exciting area as it mixes Game Systems (AI, Animation, Story engines...) with high-end 3D tools and pipeline.)

The goal: input a Story Script to a system and generate content (layout and rough 3D animation) automatically to feed a professional high-end animation pipeline for final touches and finishing.

• Teller Script stories are programmed in Python: characters, environments, actions and other performance descriptions.

• Teller reads the Script story (bottom part of the timeline in the video) and generates animation compositions using the rigged character (upper part of the timeline in the video).

• This process is performed in 3 different passes and each pass is divided in categories -i.e.; locomotion, gestures, improv, emotion, theme... (bottom part of the timeline):

  • Narrative: direct descriptions of the storyline.

  • Misenscene: additional behaviours and animations as interpreted by Teller from the story.

  • Performance: the characters in Teller act (or at least Teller adds it's own animation variations to show difference performance options based on the character state –i.e.: emotion, locomotion, gesture, improv, theme... -).

• Teller adds animation blending, transitions, time fitting and other techniques to the final animation composition.

• The final animation timeline is exported to Maya.

• This part of Teller is developed in Python/C++/Maya MEL scripts.

• Teller catalogues composable animations from Maya and tag them with behaviours annotations.

• Annotated animations can be composed dynamically in runtime to create more complex animation compositions

• Teller animation composition is exported to Maya.

• The animator can complete and polish the final animation from this first rough automated iteration.

• Animation compositions have been generated in Maya, using Teller automated generation directly from storytelling.

• The video shows multiple variations of the animation generated from different Teller passes: Narrative, "Misenscene", Performance.

Update 2012: developed another proof-of-concept with Teller approach to test a pipeline to create Adobe Flash animations using same framework and concept.

Update 2013: developed test to generate behaviour/Script nodes using in this case AWS DynamoDB, RabbitMQ, Amazon EMR. ('for fun only', so far)