July 2022 Summaries

Disco Diffusion, powered by DiscoArt on Runpod, now allows users to schedule prompts during the diffusion process, enhancing the creative potential of digital art projects. This new feature enables the alteration of prompts throughout the diffusion steps, offering greater flexibility and creativity for users. To utilize this capability, users must activate "Enable Advanced Custom Settings" within the platform, allowing them to input a JSON file with their prompt schedule. This feature supports the introduction of new prompts midway through rendering and offers the capability to toggle clip guidance for each prompt independently. Although using advanced overrides bypasses the standard UI prompts, an advanced prompt editor is being developed to improve user experience.

Runpod has introduced version 1 of its user interface for Disco Diffusion, aimed at providing a user-friendly experience while allowing users to generate a large volume of art with advanced settings. The interface features an easy-to-use settings editor, the ability to import and export settings to a text file, a queuing system for batch generation, progress updates, a gallery view, and automatic GIF generation. Runpod seeks feedback through Discord or email to enhance the user experience further, emphasizing their enthusiasm for the project and the community's involvement.

Disco Diffusion renders typically begin with an input, either Perlin noise or an initial image, and this article introduces using a Voronoi diagram as an alternative starting point. Voronoi diagrams can create intriguing visual effects by generating a noise pattern unlike the traditional TV static Perlin noise, and the number of voronoi_points can be adjusted to change the complexity of the starting pattern. Users can further customize their renders by defining static colors for Voronoi segments or using a generated palette with specified RGB ranges for more dynamic results. Although the default discoart template does not include this feature, users can manually integrate it with a Voronoi generator and modify their configuration file accordingly. Adjusting parameters such as skip_steps can accentuate the initial image in the final result, offering a variety of artistic outcomes and effects.

Google Colab Pro and Runpod offer different approaches to accessing GPU compute for data processing tasks, each with distinct advantages. Google Colab Pro and Pro+ provide an easy-to-use interface with GPU/TPU access through a subscription model, priced at $9.99 and $49.99 per month, respectively, but do not guarantee continuous compute, although Colab Pro+ offers better stability and priority access to faster GPUs. In contrast, Runpod's pay-as-you-go model offers guaranteed GPU compute with an RTX 3090 for as low as $0.30 per hour, translating to 166 hours of uninterrupted compute for $50, while allowing users control over the GPU deployment. Both services integrate with cloud storage, though Colab Pro is directly compatible with Google Drive, whereas Runpod offers flexibility with various providers. While Colab Pro sometimes provides access to multiple high-performance GPUs, Runpod ensures consistent access without interruptions, appealing to users who prefer more control over their compute resources.

Google Colab Pro and Runpod offer distinct models for accessing GPU compute resources, catering to different user preferences and needs. Google Colab Pro+ provides a subscription-based service with a user-friendly interface, offering priority access to faster GPUs and the ability for background execution, although without guaranteed compute. Despite its affordability, users may need to monitor their instances to maintain continuous operation. On the other hand, Runpod employs a pay-as-you-go model, ensuring uninterrupted use of GPUs like the RTX 3090 at competitive hourly rates, with the flexibility to choose from a variety of cloud storage providers. While Google Colab excels with integrated storage solutions like Google Drive and occasional access to high-end GPUs such as multiple A100s, Runpod offers users more control over GPU deployment and uninterrupted compute, appealing to those who prioritize reliability and flexibility in their cloud computing needs.

Runpod has introduced the option for clients to encrypt their persistent volumes, enhancing data security against cyber threats, malware, and potential breaches. While encrypted volumes offer increased data protection, they may also result in decreased performance and added management complexity. Clients can activate this feature during deployment by selecting the "Encrypt Volume" checkbox, and can either allow Runpod to manage encryption keys through its web platform or choose their own keys via API deployment. Runpod advises using encrypted volumes for sensitive data but suggests non-encrypted volumes for those who prioritize performance. For assistance or inquiries, clients are encouraged to reach out through Discord, chat, or email.