I Got Access to the FIRST AI Text to Video Generator!

TL;DR

Runway ml's Gen 2 text-to-video generator shows promise in its early alpha stages, with the potential for significant improvement by the end of 2023.

Transcript

so as some of you viewers no doubt know by now I got access to Runway ml's Gen 2 text to video generator I want to preface this this generator is still in its early Alpha stages it is not even complete yet but I was very lucky a Runway employee gave me access and I've been doing a lot of testing and I'm here today to show that to you guys I also ma... Read More

Key Insights

• 🛀 Gen 2 is still in its early alpha stages but shows potential for advancements in text-to-video generation.
• 🎮 Prompt specificity and proper understanding of video vs. image prompts are crucial for achieving accurate and coherent video generations.
• ❓ Despite limitations, Gen 2 can understand and replicate certain aspects of text prompts, such as lighting, backgrounds, and objects.
• ⛑️ Gen 2's ability to grasp more complex cinematic and action cues still requires further training and improvement.
• 🎮 Trippy, abstract, and incoherent prompts seem to be more successful in generating visually appealing videos with Gen 2.
• 🎮 Gen 2 shows signs of incorporating camera effects, such as rolling shutter and motion blur, which can add realism to video generations.
• 🎮 Gen 2's adherence to text prompts is stronger in images than videos, indicating a need for more specific video-oriented prompts.

Questions & Answers

Q: How does Runway ml's Gen 2 text-to-video generator compare to previous versions?

While still in its early alpha stages, Gen 2 shows promise and is comparable in quality to stable diffusion technology when it was first released. It is expected to make significant improvements as it continues to evolve.

Q: Can Gen 2 accurately understand and generate video based on text prompts?

Gen 2 demonstrates a partial understanding of prompts and can capture some aspects correctly, such as lighting, backgrounds, and objects. However, it still requires further training to fully grasp complex cinematic and action cues.

Q: Can Gen 2 effectively generate animated characters and scenes?

Gen 2 is capable of generating 3D animated characters and scenes. While some aspects may appear creepy or imperfect, the technology has the potential to create realistic and stylistic animations with improved training.

Q: How does Gen 2 handle nonsensical prompts?

If provided with nonsensical prompts, Gen 2 will generate imagery that matches the randomness of the input. In these cases, the resulting video may appear abstract and incoherent.

Q: How well does Gen 2 handle prompts specific to images versus videos?

Gen 2 often defaults to generating still images when prompted with image-related keywords. To achieve better video generations, it is necessary to learn how to prompt the model specifically for video output.

Q: Does Gen 2 incorporate camera effects and nuances?

Gen 2 shows signs of learning camera effects, such as rolling shutter and motion blur, when trained on specific footage types like GoPro videos. This suggests the potential for incorporating specific visual qualities in future generations.

Summary & Key Takeaways

• Runway ml's Gen 2 text-to-video generator is in its early alpha stages but shows potential for significant advancements in the future.

• Despite limitations, Gen 2 demonstrates an understanding of text prompts and can generate visually appealing video clips.

• Prompt specificity and the use of animated keywords can aid in achieving more accurate and coherent video generations.