ZastTranslate 1.21 Runs Qwen-Image-2.1 Locally: YouTube Visuals in 39s on RTX 4090

cocktailpeanut · x · 2026-09-21

ZastTranslate 1.21 is out, integrating Alibaba's Qwen-Image-2.1 (7B single-stream DiT) fully locally via Pinokio with GPU offload optimization. The demo generates a high-CTR YouTube visual in 39 seconds on an RTX 4090: reference-image identity conditioning locks the creator's face, skin tone and hairstyle inside a cinematic studio scene with crisp 3D typography — 100% local and private, no cloud subscription.

Original post →

More from Multimodal

Multimodal channel →