Krea2 merge model implementation

This is an AI text-to-image application based on the Krea2 merged model merge50, Krea2T Enhancer, and Krea2 Rebalance.

Users only need to input a prompt, and the system will use the pre-merged Krea2 model to generate ultra-vertical images. Simultaneously, it performs model-side enhancement and condition-side enhancement before sampling. Krea2T Enhancer is responsible for enhancing the texture quality of the model's output, while Krea2 Rebalance strengthens the weight of the prompt conditions. Together, they improve the artistic style and stability of the final image.

Unlike the standard Krea2 base process, this version does not directly load the original Krea2 Turbo model but instead loads merge50.safetensors. Therefore, it is more suitable for testing the visual style, character rendering, color tendencies, compositional ability, and overall texture of merged models.

The entire process is automated: loading the merged model, Krea2T Enhancer model enhancement, Qwen3VL text encoding, Krea2 Rebalance condition enhancement, 9:21 ultra-vertical image creation, batch latent creation, 20-step DDIM sampling, Qwen Image VAE decoding, image preview, and final saving.

The current default configuration is suitable for testing merged models, creating oriental fantasy posters, vertical character illustrations, artistic style artwork, mobile wallpapers, short video covers, social media visuals, and validating the fusion effects of Krea2 models.