Heartmorph: Heart-Shaped WAN2 T2V LoRA Guide

The Heartmorph LoRA for Wan2.2 T2V 14B turns fluid dynamics into a visual metaphor. Two separate elements — cream in coffee, ink on paper, ice melting on slate, ropes on a dock, or motorcycle tracks in mud — converge symmetrically to form a heart shape, rendered with cinematic overhead camera angles.

What the Model Does

Heartmorph is a text-to-video LoRA trained on top of Wan-AI/Wan2.2-T2V-A14B, a 14 billion parameter diffusion model. Instead of altering overall style or quality, it steers generation toward a specific transformation motif: the meeting of two opposing flows that resolve into a heart. The emphasis is on fluid, organic motion — billowing, swelling, curling — captured from a fixed or slowly retracting overhead perspective.

The model card includes five example prompts, each describing a pair of elements converging:

1. Cream in coffee — two streams poured from opposite sides, swirling toward the center-bottom to form a heart.
2. Motorcycles in mud — two riders carving arcs across a field, their tracks converging into a massive heart shape.
3. Ropes on a dock — two coiled ropes uncoiling and meeting at the center, their ends curling into heart lobes.
4. Melting ice on slate — two ice cubes melting, their water rivulets winding across dark slate to meet in a heart.
5. Ink on rice paper — two drops of black ink bleeding symmetrically through paper fibers to form a heart.

Trigger Words and Prompting

The trigger phrase is simply "Heartmorph". For best results, start your prompt with this word, followed by a description of the two elements that will converge. Each example prompt in the card is highly detailed — specifying the camera angle (overhead, aerial drone, extreme macro), the materials involved, and the moment of convergence. This level of detail helps the model place the heart formation precisely.

The prompts also emphasize that the camera remains largely stationary, watching the transformation unfold rather than cutting or panning. This restraint is what gives the results their hypnotic, meditative quality.

How to Use in ComfyUI

To use Heartmorph in ComfyUI:

1. Download [WAN2.2]Heartmorph-Redmond_high_noise.safetensors and place it in ComfyUI/models/loras/.
2. Ensure your environment is configured to run Wan models.
3. In your workflow, load your base Wan model.
4. Add a Load LoRA node and connect the base model output to its model input.
5. Select the Heartmorph safetensors file in the Load LoRA node.
6. Connect the MODEL output of the LoRA node to your KSampler.
7. Set strength_model between 0.8 and 1.0 to start; adjust to taste.

The model is loaded as a high-noise LoRA, meaning it has the greatest effect in the early denoising steps. This makes it well-suited for controlling the overall composition rather than fine detail.

Recommended Settings

There are no published resolution, CFG scale, or step-count recommendations beyond the LoRA strength range. Because the model operates on the Wan2.2 14B base — which supports 1280x720 generation at 16 fps for short video sequences — matching those defaults and setting CFG around 5-7 typically produces stable results. The key variable is the LoRA strength, which controls how strongly the heart-morphing motif appears.

License

Heartmorph is released under CC0-1.0 (Creative Commons Zero), placing it in the public domain. It can be used freely for commercial and non-commercial projects without attribution, though crediting the creator and reDMOND Ai is appreciated.

Sources

• Hugging Face model card — trigger words, example prompts, ComfyUI setup, base model reference.
• Wan2.2-T2V-A14B base model — the 14B text-to-video model Heartmorph adapts.