godwenbin

> Hi, tool_add_anytext.py 是用来生成一个用作从头训练的anytext ckpt，这里面的weights都是随机初始化或copy自sd1.5的，所以不能生成文字。只有demo.py有使1.5系列的base model或lora生成文字的功能，可以将合并weight后的ckpt保存下来。 感谢回复！请问下demo.py为什么能使我的1.5系列基模具有生成文字的能力呢？然后您提到合并后的ckpt保存下来，我目前是跑了tool_add_anytext.py生成了一个合并weight的模型，如您所说，这是一个用作从头训练的anytext ckpt，我放到webui里面确实不具备生文字能力。想请问下，有更快捷的方法让我的1.5系列基模拥有anytext的能力吗？ 另外，我看readme里面写的Change base model: Simply fill in your local base model's path in the [Base Model Path].在demo更换我的基模，这边是能够评估基模的哪一方面能力呢？

> Some features for VRAM optimization do not work very well with multi-GPU setups, such as --fused_backward_pass. However, 80GB of VRAM should at least work with a batch size of...

angthing update?

> Could please upload the image again? it's just a pure noise image after i try to finetune the flux. and i think may be my finetune code 's error....

Could you please tell me which UI or demo scripts you used when inferring the flux fine-tuned model?