This Img2Prompt2Img workflow for KREA is for those situations where you want to make sure that you like the concept of an image but not it's implementation. Say you have a semi-realistic image that you want to make it real without actually directly using the image, in such cases you need a prompt to get core elements of the referenced image to generate your new Realistic output, this automates the process. No LORAs, No Style Transfers, No complex spider web of node connections. Using this is very simple -
#1. Load KREA and Florance 2 models
#2. Load your Reference Image for automated prompt generation
#3. Set your output size and click Run
...That's it.
================================================================================================
You need to download your necessary models from CivitAI and using ComfyUI Manager. These are the models you will need -
Flux Models -
https://huggingface.co/QuantStack/FLUX.1-Krea-dev-GGUF/resolve/main/flux1-krea-dev-Q4_0.gguf
or
Flux Encoders -
https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/clip_l.safetensors
https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/t5xxl_fp8_e4m3fn.safetensors
Flux VAE -
https://huggingface.co/black-forest-labs/FLUX.1-dev/resolve/main/ae.safetensors
If you are missing any nodes use ComfyUI Manager to get them, and only on your first ever run use Florance 2 downloader and loader (move node connection from regular Florance 2 Model Loader) otherwise use the regular one. Enable that one on very first (make sure to disable the regular loader) run it will get and set Florance 2 model to the correct path, on your second run you can get back to regular one. If you are interested to try out other workflows ( for HiDream, WAN, QWEN Image and Flux ) I also included a text file on the archive that has this workflow file, in which you will find even more links for essential HuggingFace downloads for my other workflows I have on my CivitAI profile.