🍆 CoachBate — Qwen Image 2.1 Penis LoRA

Coach keeps bringing quality D to the latest image models. Today, it’s Qwen Image 2.1—with native 2K generation and image editing in the same model. This is a brand new model released September 2026

This is a non-commercial model, so it's free. I most likely won't be updating it since it is hard to train.

Generate a new image from a prompt, or work with an image you already have. Qwen Image 2.1 combines both workflows, giving this release applications beyond text-to-image alone. This training dataset contained soft and hard circumcised penises and cumshots so it knows both of these topics. There is a separate uncircumcised penis lora.

Check the showcase for example results and the prompts used to create them. They’re a useful starting point for seeing how to phrase your own prompts.

⚙️ Settings: check your CFG

My recommendation for this LoRA’s text-to-image results is CFG 3.

The ComfyUI template uses CFG 1. In my testing, that produced poor results with this LoRA; changing it to 3 made a substantial difference.

For context, the official text-to-image template starts with Euler, the simple scheduler, and 25 steps. Its negative prompt is unused at CFG 1. These are template settings, separate from my CFG recommendation. Official text-to-image workflow

🔎 Optional anatomical vocabulary

You don’t need to include anatomical terminology in every prompt. These words are optional and can be used for more control when you want to describe particular structures:

• Glans

• Shaft

• Frenular delta

• Urethral orifice (meatus)

Keep your prompt simple when it’s giving you the results you want. Add specific terms when you need to direct attention toward a particular anatomical region or detail. See the showcase prompts for examples.

🖼️ Image editing and references

The editing side lets you describe changes to an existing image, including anatomical proportions and detail, while requesting preservation of the surrounding subject and composition.

Qwen Image 2.1 supports up to 10 reference images. In the official ComfyUI editing workflow:

• <image1> is the image being edited.

• Additional images supply references.

• Identify their roles using <image1>, <image2>, and so on.

• Describe the requested change and the features you want preserved.

With custom sizing disabled, the output canvas follows the processed first image. Additional references can have different dimensions and aspect ratios. Official editing workflow

All current showcase images are text-to-image generations. They demonstrate that side of the LoRA; editing performance hasn’t yet been established by the showcase.

📐 Native 2K resolution

Qwen Image 2.1 supports direct generation at approximately 4 megapixels, including 2048 × 2048 square images - see Official resolution guidance

💬 Show Coach what you make

Share your generations and edits, along with your prompt, LoRA strength, CFG, sampler, steps and output dimensions using the Add Post below.

For editing, before-and-after comparisons are especially useful: they show the anatomical changes and how well the original subject and scene were preserved.