First Pass — Base Image (Qwen Image Model)
Loads the primary diffusion model, CLIP, and VAE.
Uses a Qwen Master Prompt to generate the initial semantic prompt.
Converts the text prompt + negative prompt into CLIP conditioning.
Sets the image size via an Empty Latent node.
Runs a KSampler to generate the first-pass image.
Decodes the latent to an actual image.
Sends this image to a preview and to the comparison block.
2. Second Pass — Refinement (Z-Turbo Model)
Loads a second, faster refinement model (Z-Turbo) along with matching CLIP and VAE.
Reuses the first pass image by re-encoding it back to latent form.
Adds additional positive/negative text conditioning.
Runs a second KSampler tuned for enhancement.
Decodes the refined latent to the final image.
Previews the refined image and compares it with the first-pass output.
3. overall goal
Generate an initial image using a high-quality Qwen model.
Feed that image into a second model (Z-Turbo) for cleanup, sharpening, and stylistic improvement.
Allow visual comparison between the initial and refined outputs.