Mimikyu Costume Generator

The pipeline uses a combination of ControlNet (Canny) for structural guidance and a Mimikyu LoRA for character consistency. It follows a "Reference → Silhouette Extraction → Character Injection" logic to ensure the final result looks like a crude, handmade disguise rather than a generic digital filter.

Reference Ingestion & Silhouette Extraction:

• LoadImage (Node 9): Imports the subject you want to turn into a costume.

• CannyEdgeDetect (Node 11): Extracts the specific angular contours and geometric shapes of the subject. This is crucial for capturing the unique features (like the hexagonal eyes or helmet silhouettes) you requested.

• Character & Style Guidance:

• LoraLoader (Node 2): Injects the Mimikyu LoRA. This ensures the creature's anatomy (hunched body, stick tail) remains intact regardless of the costume.

• CLIPTextEncode (Node 3/4): Contains the specialized "Handmade/Patchwork" prompt. It forces the model to interpret the ControlNet shapes as cloth, paper, and stitches instead of metal or skin.

• Structural Control:

• ControlNetApply (Node 12): Merges the Canny edges with your text prompt. It "pins" the costume's shape to the reference image while allowing the diffusion model to fill in the "creepy-cute" patchwork textures.

• Core Generation:

• KSampler (Node 6): The engine that renders the image. It is tuned for a high ⁠denoise (1.0) to ensure a complete transformation from the reference image into the stylized anime/concept-art aesthetic.