Kian (@kian_xyz) 's Twitter Profile
Kian

@kian_xyz

Building @koro_world_ |
Exploring the latent space |
Node-based cowboy

ID: 1539362952867893248

calendar_today21-06-2022 21:43:15

492 Tweet

198 Followers

745 Following

Ostris (@ostrisai) 's Twitter Profile Photo

You can now train LoRAs for Qwen Qwen Image Edit 2511 with AI Toolkit. I also trained a 3bit Accuracy Recovery Adapter for it that will allow you to finetune it at 3bit with <24GB of VRAM.

You can now train LoRAs for <a href="/Alibaba_Qwen/">Qwen</a> Qwen Image Edit 2511 with AI Toolkit. I also trained a 3bit Accuracy Recovery Adapter for it that will allow you to finetune it at 3bit with &lt;24GB of VRAM.
Ostris (@ostrisai) 's Twitter Profile Photo

I am testing training a LoRA using student teacher training with Z-Image (base) as the student and Z-image-Turbo as the teacher to make a base turbo LoRA. These samples are 5k steps at a batch of 1. Sample is 8 steps CFG 1. Before -> After. I am honestly shocked it worked.

I am testing training a LoRA using student teacher training with Z-Image (base) as the student and Z-image-Turbo as the teacher to make a base turbo LoRA. These samples are 5k steps at a batch of 1. Sample is 8 steps CFG 1. Before -&gt; After. I am honestly shocked it worked.