Spaces:

---

akhaliq

/

HunyuanImage-3.0

like 175

Running

If you don't mind generating image of higher quality, then would be great if you return 2048x2048 and PNG (currently you return WEBP).

I don't know for sure but to my mind you need to change:

app.py file, by adding format = 'png', type = 'pil', to params of output_image = gr.Image( call.

models.py file, by adding width = 2048, height = 2048, to params of image = CLIENT.text_to_image( call.

If for some reason FAL-AI returns not PNG then you can add extra_body = dict(output_format = 'png'), to .text_to_image( (mentioned above).

Also you can experiment by adding num_inference_steps = 40, to .text_to_image(, by default FAL-AI has 28 inference steps for model Hunyuan.