Key insights

• ⚡ Spell Rush uses deep learning to create AI tools for art and artists, allowing for scalable content creation without massive studio pipelines and high budgets.
• 🎨 Their AI tools can create images on par with what a human illustrator would produce, reducing time and cost for content creation.
• 🤖 They have developed a tool using generative adversarial networks (GANs) to create characters in seconds, enabling customization and mass production of characters.
• 🌌 Spell Rush's process of training a generator using GANs can produce high-quality, indistinguishable images from real data, reducing the time for classical painting creation.
• 🎭 Their AI can efficiently generate multiple character variations, with efforts to improve representation of darker skin tones and male characters in the dataset.
• 👫 AI tools enable the creation of high-quality male characters, addressing the preference for female characters by illustrators in Japan, and automated animation in a cost-effective manner.
• 💼 The company uses a DIY supercomputer for AI training, which costs $3000-$4000 per model and is looking to hire for various positions including artists and AI research intern.
• 🕹 They are running crisis-capable AI workloads on a cluster of GPUs to develop the world's first AI-illustrated game and are hiring for art and AI research positions.

Q&A

• What are some positions Spell Rush is hiring for?

  Spell Rush is looking to hire for various positions including 2D animator, motion designer, real-time VFX artist, and AI research intern. Interested candidates can contact jobs@swellbrush.com.

• What kind of hardware does Spell Rush use for running its AI workloads?

  Spell Rush is a small startup using a cluster of GPUs, including Titan RTX GPUs, to run crisis-capable AI workloads at a low cost.

• How is Spell Rush training its AI models and what are the associated costs?

  Training the models is expensive and takes 7-10 days, costing $3000-$4000 per model due to time and resource requirements.

• Why does Spell Rush use AI for drawing male characters?

  In Japan, illustrators prefer drawing female characters for higher social media engagement, but AI can be used to draw high-quality male characters. The company also uses AI for automated animation and workflow tools, trained on a DIY supercomputer due to cost considerations.

• What kind of dataset does Spell Rush's AI use for generating characters?

  The dataset includes 10 million images mainly from the anime aesthetic, with efforts made to improve the representation of darker skin tones and male characters.

• What is the process of training a generator for creating classical paintings?

  The process involves random image evaluation by a discriminator, followed by weight updates for both the generator and discriminator. The generator requires noise input to create diverse images and can eventually produce high-quality images indistinguishable from real data.

• What kind of AI tools has Spell Rush developed?

  Spell Rush has developed AI tools that can create characters in seconds, enable customization and mass production of characters, and even produce high-quality, realistic images indistinguishable from real data.

• How does Spell Rush's AI tools compare to human illustrators?

  Spell Rush has developed AI tools that can create images on par with what a human illustrator would produce, enabling efficient and cost-effective content creation.

• What is Spell Rush using deep learning for?

  Spell Rush is using deep learning to create AI tools for art and artists, allowing them to scale up content creation without massive studios and high budgets.

• 00:02 Spell Rush is using deep learning to create AI tools for art and artists, making it possible to scale up content creation without massive studios and high budgets. They have developed AI tools that can create images on par with what an illustrator would produce.
• 01:44 A tool can create characters in seconds using generative adversarial networks (GANs), enabling customization and mass production of characters.
• 03:20 A process of training a generator to produce classical paintings through random image evaluation by a discriminator, followed by weight updates for both generator and discriminator. The generator needs noise input to create diverse images and can eventually produce high-quality images indistinguishable from real data.
• 05:00 AI can generate multiple character expressions, colors, and styles quickly. Dataset includes 10 million images mainly from the anime aesthetic. Efforts made to improve representation of darker skin tones and male characters.
• 06:48 Illustrators in Japan prefer drawing female characters for more likes and retweets, but AI can be used to draw high-quality male characters. The company uses AI for automated animation, trained on a DIY supercomputer due to cost. Training the models is expensive and takes 7-10 days, costing $3000-$4000 per model.
• 08:27 A small startup with a cluster of GPUs running crisis capable AI workloads at a low cost. Developing an AI illustrated game and hiring for various positions including artists and AI research intern.