OPEN-BANANA-REMIX

This is a remix version of a context model that utilizes the feature of QWEN edit to read the style lora of QWEN image. QWEN edit is used as high noise, and kontext as low noise. The image generation follows a three-stage denoising process.

It supports composition, high-noise stylization, and low-noise stylization with three loras.

1) Generally speaking, switching the high-noise stylization lora alone can achieve good results. Adjust the weight if necessary.

2) Our own style painting model is not yet trained. You can temporarily use other QWEN models from the open-source community instead.

3) The higher the value for low-noise denoising, the closer the image will resemble the default style of the low-noise model (kontext). Try not to set it too high. Generally, values between 0.6 and 0.7 are acceptable.

This solution provides relatively flexible style-switching capabilities. Since the low-noise model is included, the image quality is higher than that of JiMeng 4. However, it seems that prompt adherence is slightly lower than JiMeng at the moment.