Post by DeepInfra

We just added text-to-video on DeepInfra → and it's fast. FastWan-QAD-FP8-1.3B from FastVideo (Hao AI Lab) generates 5-second 480p video clips from a text prompt in about 2.4 seconds. Landscape or portrait. $0.0025 per second of video. A few things that make this model interesting: → 3-step inference via quantization-aware distillation — dramatically fewer steps than standard diffusion models → FP8 quantized linear layers with SageAttention2++ backend — fast on Ampere and Ada GPUs, not just the latest hardware → No classifier-free guidance at inference — simpler, faster decoding → Fast decoding via TAEHV tiny autoencoder At $0.0025/second, a 5-second clip costs $0.0125. That's cheap enough to generate video at scale — product demos, social content, creative tools, anything where you'd normally rely on expensive commercial video APIs. Available now on DeepInfra via our standard API. → https://lnkd.in/gTgczbYV