Sure! Here’s the translation into American English:
A new approach to creating animated storyboards has emerged thanks to the use of artificial intelligence, allowing for notable visual consistency among characters in audiovisual productions. This technique, which is based on image engineering and character development, enables creators to fine-tune AI models, specifically the Amazon Nova Canvas, to accurately manage the appearances and expressions of characters across different scenes.
FuzzyPixel, a division of Amazon Web Services (AWS), has implemented an innovative project using the animated short Picchu as a foundation. By extracting keyframes, training data is prepared that maintains the consistency of the main characters, such as Mayu and her mother, facilitating the rapid generation of storyboard concepts for future sequels.
The automated workflow begins with uploading a video asset to an Amazon Simple Storage Service (S3) bucket. This process encompasses multiple stages, including reducing the resolution of the frames and selecting those that show the characters. Additionally, theAmazon Nova model is used to generate subtitles, further enriching the content.
The character extraction involves capturing video frames at fixed intervals, addressing label detection and face recognition to identify the characters. This activity is complemented by a deduplication algorithm that ensures the diversity of the dataset by removing similar images that could lead to model overfitting.
Once sufficient labeled images have been collected, the data quality is verified through a human-in-the-loop process, ensuring that only accurate information is used for training. Positive results in preliminary tests suggest that correct adjustments to hyperparameters can lead to significant improvements in visual consistency.
After fine-tuning, the model is made available for deployment, which can be done from the Amazon Bedrock console or using the Python SDK for more customized integration. This flexibility allows creators to test the model to generate new images, maintaining stylistic and quality consistency in their storytelling.
With this innovative methodology, a substantial acceleration instoryboard production is anticipated, along with an elevation in the quality of visual content. This will allow creative teams to focus more on narrative and less on technical consistency.
Source: MiMub in Spanish
