Workflow Using Krea in nunchaku version or QWEN. T2I, I2I, Creaprompt, which allows you to create prompts from thematic lists. Ollama, which allows you to use 7 prompt systems:

- Photo prompt: give a word or a short phrase , an idea, and it creates a prompt.

- Surrealism: The same thing but oriented towards surrealism.

- Character style: Give a character with a location and specify the style (it can be anything). A prompt will be created with all the elements of the image to which the style will be applied.

- Prompt variation: Enter a prompt and a variation of that prompt will be created. The general structure of the prompt is retained, but some elements change.

- Image prompt: Upload an image, set a model vision for Ollama, and a prompt is created describing the image.

- Enhanced prompt: Very effective if you provide a concise prompt of 3 or 4 lines with the essentials. The LLM will add details to all elements and may even add other elements.

- Prompt random : Create a random prompt: There must be instructions in ollama generate: It is better to use a prompt without any visuals, only improvement terms. You can use the one in the note alongside. You can also guide it by adding a few elements in ollama generate.

A NODE: ‘MultiSwitchText.py’ MUST BE PLACED IN THE ‘custom_nodes’ DIRECTORY.

v1.1 : updated to work with the latest version of nunchaku.

v1.5 : New version of the workflow with the following main changes:

- New prompt system in Ollama: random image

- Integration of QWEN realistic version with which you can use Ollama, Creaprompt, or a normal text prompt.

- New 2-pass upscaling system for Krea: with an SDXL model that refines the image, a downscaling, and finally a new upscaling with KREA.

v1.6 :

- Centralization of the system for loading images: with the option to open a single image or a batch of multiple images in a directory.

- Use of Qwen3-vl node in addition to ollama for llm.

- Ability to save the prompt created by llm.

- Ability to use creaprompt with ollama.

- 2-pass upscaling system for Krea and now for Qwen.

v1.7:

- Some modifications to certain prompt systems.

- Some adjustments for upscaling.

- Replacement of qwen3-vl nodes with minicpm nodes, which are more efficient and use fewer resources. Creation of a prompt from three images: 1) image to be reproduced 2) image that provides the style 3) image that provides the color palette and lighting.