--- tags: - text-to-image - lora - diffusers - template:diffusion-lora widget: - text: >- pixel art, a rainy day scene of a cat-eared anime girl with a transparent umbrella, standing at a train crossing, her reflection shimmering in the puddles as a train rushes by. output: url: images/ComfyUI_00021_.webp - text: '-' output: url: images/ComfyUI_00021_.webp base_model: tencent/HunyuanVideo instance_prompt: pixel art license: mit --- # Hunyuan Video Lora - PixelArt ## Model description # Hunyuan Video LoRA - PixelArt **(Model card WIP; subject to updates today or tomorrow)** [v1.0]: This LoRA brings an **anime-inspired pixel art style** to life, trained on a mix of pixel animations and still images. It’s designed to generate vibrant, colorful anime-style pixel art, excelling at **character motions** and **pixelated scenery**. Think bright, dynamic visuals with that classic retro charm. ## Usage A sample workflow is available in the Hugging Face repo: -[ [v0.1/ComfyUI_00024_.webp · trojblue/HunyuanVideo-lora-AnimeShot\]](https://huggingface.co/trojblue/HunyuanVideo-lora-AnimeShot/blob/main/v0.1/ComfyUI_00024_.webp) \- For default node compatibility, update ComfyUI to the latest commit. More details here: [[HunyuanVideo Native Support in ComfyUI - by Jo Zhang\]](https://blog.comfy.org/p/hunyuanvideo-native-support-in-comfyui). ## Configs Trained on a dynamic aspect ratio at **~768 resolution**, this LoRA has been tested at resolutions: - 768x768 - 656x880 - 880x656 It’s flexible with step counts, working effectively at: - 1 step - 33 steps - 65 steps - 97 steps Feel free to tweak these settings to find your sweet spot! ## Prompting Start prompts with `pixel art, <description>`. The training data used **natural language captions** (1-3 sentences), so descriptive, sentence-style prompts tend to yield the best results. Here are the prompts behind the sample images: ``` pixel art, a pixelated scene showing a fox-eared anime boy with fiery red hair and golden eyes, sitting on a torii gate at sunset, gently playing a bamboo flute as cherry blossoms float in the air. pixel art, a pixelated image of a silver-haired shrine maiden with glowing violet eyes, sweeping the temple courtyard under the full moon, with soft lantern light and petals drifting by. pixel art, a pixel scene of a pink-haired demon loli with tiny horns, roasting marshmallows over a campfire in a haunted forest, surrounded by glowing ghost friends. pixel art, a rainy day scene of a cat-eared anime girl with a transparent umbrella, standing at a train crossing, her reflection shimmering in the puddles as a train rushes by. pixel art, a rainy day scene of a cat-eared anime girl with a transparent umbrella, standing at a train crossing, her reflection shimmering in the puddles as a train rushes by. pixel art, a pixelated mountaintop at dawn with a lone boy in a scarf standing at the edge, arms outstretched as the wind carries cherry petals past him, a red kite fluttering behind. pixel art, a pixelated scene of a frog in a wizard hat stirring a bubbling cauldron in the middle of a mushroom forest, fireflies glowing around him like sparks of magic. pixel art, a pixelated alleyway in a quiet neon-lit city, a boy with silver hair feeding a stray black cat from a bento box, both bathed in soft vending machine light. pixel art, a peaceful farm landscape with a scarecrow slightly askew in the middle of a golden wheat field, crows perched on its arms as the wind moves the grain in waves. pixel art, a quiet underwater scene where a catfish wearing a crown floats lazily above a coral throne, tiny sea creatures circling like royal attendants. ``` ## Limitations The dataset leans heavily on **anime characters and scenery**, so prompts outside this scope might produce less pixelated or lower-quality results. I may experiment with WAN to improve this later, but for now, this is the Hunyuan version as it stands. Training wrapped up around January—I just didn’t get around to posting it sooner (my bad!). As a result, the inference setup hasn’t been updated to the latest Hunyuan Video best practices. Checking recent guidelines could help optimize your results. ## Updates feel free to follow me on twitter for model updates and stuff: [[yada (@yada_cc) / X\]](https://x.com/yada_cc) ## Trigger words You should use `pixel art` to trigger the image generation. ## Download model Weights for this model are available in Safetensors format. [Download](/trojblue/HunyuanVideo-lora-PixelArt/tree/main) them in the Files & versions tab.