Jin's Daily AI Trivia: First Usable Open-Source AI Video Generator (FLF2V)?

Jin’s Daily AI Trivia: First Usable Open-Source AI Video Generator (FLF2V)?

Alibaba (the Qwen gang) just dropped their first 14B First-Last-Frame-to-Video (FLF2V) generation model — and it’s capable of producing super smooth 720P videos.

The FLF2V method is delivering some of the best results we’ve seen so far, and it even supports text instructions for fine-tuning the output.

Wan2.1 is one of the few open-source T2V/I2V video generators that you can actually run on consumer-grade GPUs (thanks to FP8 quantization), and Alibaba has been absolutely killing it since February 2025.

Their T2V-1.3B model only needs 8.19 GB of VRAM, meaning it works on almost all consumer GPUs. It can generate a 5-second 480P video on an RTX 4090 in around 4 minutes — and that’s without using any optimization tricks like quantization. Performance-wise, it’s even giving some closed-source models a run for their money.

I’m planning to wait for the quantized GGUF version and then test it on Apple Silicon. Right now, the 14B model needs more than 40GB VRAM to run.

Meanwhile, you can use their Wan Video platform to generate your video. (https://wan.video/)

Hope you learn something new today, see ya!