We are using two images. The first image goes into the CLIP Vision Encode node, then Apply Style Model node with Flux Redux model.

Then we are connecting Apply Style Model's conditioning node to Conditioning(combine) node.

For the second image, I'm using LLaVA-OneVision Run node, which is using LLava-OneVision VLLM model. Goal is extracting a detailed prompt from our image. My instruct prompt is also very powerful for extracting the details.

You can use any VLLM node such as Ollama Vision, or totally remove the LLM Nodes, and write your own prompt into the CLIPTextEncodeFlux node. If you want to manually enter the prompt, remember to right click into the CLIPTextEncodeFlux node > Convert Input to Widget > Convert t5xxl to widget.

The CLIPTextEncodeFlux node's output conditioning connects into the second input node of the Conditioning(combine) node, but also it connects into the ApplyStyleModel node's conditioning node.

In between, I'm using 5 LoRA's for the styles, some of them are mine some of them are public LoRA's. You can use any lora you want, but my image styles mostly coming from them.

I'm also using Detail Daemon custom node for extra details in my images.

LoRA's:

1- Flux Ancient Style (my own lora):https://civitai.com/models/715253

2- Flux Ancient Style v2: My own lora again, but didn't published yet.

3- Flux Cinematic Portraits: My own lora, but didn't published yet, useful on mostly human subjects.

4- Flux Epic Gorgeous Details: https://civitai.com/models/736706?modelVersionId=863655

5- Flux Realism by Xlabs: https://huggingface.co/XLabs-AI/flux-RealismLora