--- tags: - text-to-image - lora - diffusers - template:diffusion-lora widget: - output: url: images/Boogu_image_edit_00079_.png text: >- A pixel art spritesheet of a witch with long red hair and blue eyes, wearing a purple hat and robes trimmed with white and light purple colors. The spritesheet is a 6 by 4 grid of four rows and six columns of frames - First row is 3 walking frames facing down, 2 arm swing attack frames facing down, and 1 hurt frame facing down. Second row is 3 walking frames facing left, 2 arm swing attack frames facing left, and 1 hurt frame facing left. Third row is 3 walking frames facing right, 2 arm swing attack frames facing right, and 1 hurt frame facing right. Fourth row is 3 walking frames back view facing up, 2 arm swing attack frames back view facing up, and 1 hurt frame back view facing up. - output: url: images/Boogu_image_edit_00080_.png text: >- A pixel art spritesheet of a medieval knight wearing metal armor and a helmet with a red plume. The spritesheet is a 6 by 4 grid of four rows and six columns of frames - First row is 3 walking frames facing down, 2 arm swing attack frames facing down, and 1 hurt frame facing down. Second row is 3 walking frames facing left, 2 arm swing attack frames facing left, and 1 hurt frame facing left. Third row is 3 walking frames facing right, 2 arm swing attack frames facing right, and 1 hurt frame facing right. Fourth row is 3 walking frames back view facing up, 2 arm swing attack frames back view facing up, and 1 hurt frame back view facing up. - output: url: images/Boogu_image_edit_00081_.png text: >- A pixel art spritesheet of a werewolf with light gray fur and yellow eyes, wearing a red scarf around his neck, and brown leather pants. The spritesheet is a 6 by 4 grid of four rows and six columns of frames - First row is 3 walking frames facing down, 2 arm swing attack frames facing down, and 1 hurt frame facing down. Second row is 3 walking frames facing left, 2 arm swing attack frames facing left, and 1 hurt frame facing left. Third row is 3 walking frames facing right, 2 arm swing attack frames facing right, and 1 hurt frame facing right. Fourth row is 3 walking frames back view facing up, 2 arm swing attack frames back view facing up, and 1 hurt frame back view facing up. base_model: Boogu/Boogu-Image-0.1-Edit instance_prompt: null license: apache-2.0 --- # pixel_spritesheet_4walk_combat_32x48_v2 ## Model description A pixel art spritesheet LoRA for small 32x48 characters, with animation frames for walking up/down/left/right, 2 attack frames with an arm swing, and 1 hurt frame. Trained on Boogu Image 0.1 Edit. ## How to use Use `boogu_image_edit_turbo_hotfix_1k_20260708_int8_convrot.safetensors` from Comfy-Org's [Hugging Face repo](https://huggingface.co/Comfy-Org/Boogu-Image) You'll also need the custom node `TextEncodeBooguEditAnyRes` from [here](https://github.com/svntax/ComfyUI-BooguImageEdit-AnyRes) Drag and drop the sample images into ComfyUI for the workflow, or use the one in the repo. Image outputs should be 768x768 because that's the size of the spritesheets used in training. But, you can use the model as a normal edit model or text-to-image too. ### Why a custom node? ComfyUI's default `TextEncodeBooguEdit` node has issues with some resolutions, where you don't get an output matching your reference image's size. The custom node fixes this. ## How to get pixel-perfect images Downscale by a factor of 4. So 768x768 images should downscale to 192x192. The workflow uses [PixelOE](https://github.com/KohakuBlueleaf/PixelOE) with k-centroid mode, but other 4x downscale can work too. # Using an image reference Prompt: *Create a pixel art spritesheet of the character in the image. Keep the appearance the same. The spritesheet is a 6 by 4 grid of four rows and six columns of frames - First row is 3 walking frames facing down, 2 arm swing attack frames facing down, and 1 hurt frame facing down. Second row is 3 walking frames facing left, 2 arm swing attack frames facing left, and 1 hurt frame facing left. Third row is 3 walking frames facing right, 2 arm swing attack frames facing right, and 1 hurt frame facing right. Fourth row is 3 walking frames back view facing up, 2 arm swing attack frames back view facing up, and 1 hurt frame back view facing up.* ![demon_man_image_ref](https://cdn-uploads.huggingface.co/production/uploads/68dcbc0eb3e9381d15e2cbbc/6pSMAQiaINJDUyrLeQrPq.png) ![astronaut_image_ref](https://cdn-uploads.huggingface.co/production/uploads/68dcbc0eb3e9381d15e2cbbc/qaJPnBUJyQ-rT-cBjYMdN.png) ![bluejay_image_ref](https://cdn-uploads.huggingface.co/production/uploads/68dcbc0eb3e9381d15e2cbbc/jxUg8dJ9rQwgD9bxq_9Qz.png) ## Image to pixel art Surprisingly, you can use this LoRA to change a reference image into pixel art, but it doesn't work with every type of image, and seems to work best at 768x768 and similar scale. Example: *Change this image into pixel art* ![img_to_pixel_5](https://cdn-uploads.huggingface.co/production/uploads/68dcbc0eb3e9381d15e2cbbc/1N1WrSnjrp9qn3b-BiyZN.png) ## Notes & Issues - Backgrounds from the reference image carry over to the spritesheet, so if you have a reference image with a background, add something like "White background" to the prompt so the spritesheet doesn't generate it. - Normal edit prompts still have the same weaknesses that the original model has (character identity drift, complex pose changes). ## Download model [Download](/svntax-dev/pixel_spritesheet_4walk_combat_32x48_v2/tree/main) them in the Files & versions tab.