(This template is used for Workflow Contest)

What this workflow does

👉 Input text prompt and creates a 720x1280 base image that is then tile upscaled and then upscaled more to about 4k.

How to use this workflow

👉 Input text prompt for positive and negative, input any specific face specifications in the positive conditioning inputs for face details, then click Queue Prompt.

Tips about this workflow

👉 Image generation takes approximately one and a half minutes on a 3090 Gfx Card.

Key points are:

1. Dynamic Thresholding to use a large CFG on initial image generation.

2. Kohya Deep Shrink for a larger base image creation with SD1.5 checkpoint models.

3. Using a KSampler with Restarts for improved sampler results.

4. Tiled Controlnet for image up-scaling (if low on VRAM you can halve or quarter the tile width and height       and would also need to add padding for seams and then increase the Kohya downscale by the same   proportion this is currently set to use

   one giant tile to avoid seams).

5. 4x upscaling then down-scaling to close to 4k resolution.

6. A different checkpoint is used for base image generation and up-scaling the image for potentially more interesting results.

🎥 Video demo link (optional)

👉 [Please add here]