import spaces import gradio as gr import numpy as np import torch from toonmage import attention_processor as attention from toonmage.pipeline import ToonMagePipeline from toonmage.utils import resize_numpy_image_long, seed_everything torch.set_grad_enabled(False) pipeline = ToonMagePipeline() # other params DEFAULT_NEGATIVE_PROMPT = ( 'cross-eyed, blurry, deformed eyeballs, deformed, deformed or partially rendered eyes, partially rendered objects, low resolution, disfigured hands, ugly, mutated, glitch,' 'watermark, text, artifacts noise, worst quality, low quality, non-HDRi, lowres, flaws, flaws in the face, flaws in the eyes, extra limbs, signature' ) @spaces.GPU def run(*args): id_image = args[0] supp_images = args[1] prompt, neg_prompt, id_scale, mode, id_mix, steps, seed, n_samples, scale, H, W = args[2:] pipeline.debug_img_list = [] if mode == 'fidelity': attention.NUM_ZERO = 8 attention.ORTHO = False attention.ORTHO_v2 = True elif mode == 'extremely style': attention.NUM_ZERO = 16 attention.ORTHO = True attention.ORTHO_v2 = False else: raise ValueError if id_image is not None: id_image = resize_numpy_image_long(id_image, 1024) id_embeddings = pipeline.get_id_embedding(id_image) if supp_images is not None: sup_images = resize_numpy_image_long(supp_images, 1024) supp_id_embeddings = pipeline.get_id_embedding(supp_images) id_embeddings = torch.cat( (id_embeddings, supp_id_embeddings if id_mix else supp_id_embeddings[:, :5]), dim=1 ) else: id_embeddings = None seed_everything(seed) ims = [] for _ in range(n_samples): img = pipeline.inference(prompt, (1, H, W), neg_prompt, id_embeddings, id_scale, scale, steps)[0] ims.append(np.array(img)) return ims, pipeline.debug_img_list _MARKDOWN_ = """

This is the demo for ToonMagev2 for Image to Image Stylization

**Tips** - Some examples are provided at the botton, we recommend to try the example prompts first - a single ID image is usually sufficient, you can also supplement with additional auxiliary images - Two modes are offered in this space, i.e. fidelity mode and extremely style mode. In most cases, the default fidelity mode should suffice. If you find that the generated results are not stylized enough, you can choose the extremely style mode. Try out with different prompts using your image and do provide your feedback. Note: The image of the women is generated using AI, therefore it does not resemble any living person. **Demo by [Sunder Ali Khowaja](https://sander-ali.github.io) - [X](https://x.com/SunderAKhowaja) -[Github](https://github.com/sander-ali) -[Hugging Face](https://huggingface.co/SunderAli17)** """ theme = gr.themes.Soft( primary_hue="blue", secondary_hue="purple", font=[gr.themes.GoogleFont('Source Code Pro'), gr.themes.GoogleFont('Public Sans'), 'system-ui', 'sans-serif'], ) # js_func = """ # function refresh() { # const url = new URL(window.location); # if (url.searchParams.get('__theme') !== 'dark') { # url.searchParams.set('__theme', 'dark'); # window.location.href = url.href; # } # } # """ with gr.Blocks(title="ToonMagev2", theme = "Yntec/HaleyCH_Theme_Orange") as SAK: gr.Markdown(_MARKDOWN_) with gr.Row(): with gr.Column(): submit = gr.Button("Generate") with gr.Row(): face_image = gr.Image(label="ID image (main)", sources="upload", type="numpy", height=256) supp_image1 = gr.Image( label="Additional ID image (auxiliary)", sources="upload", type="numpy", height=256 ) prompt = gr.Textbox(label="Prompt", value='portrait,cinematic,wolf ears,white hair') neg_prompt = gr.Textbox(label="Negative Prompt", value=DEFAULT_NEGATIVE_PROMPT) with gr.Row(): id_scale = gr.Slider(label="ID scale", minimum=0, maximum=5, step=0.05, value=0.8, interactive=True) mode = gr.Dropdown(label="mode", choices=['fidelity', 'extremely style'], value='fidelity') id_mix = gr.Checkbox( label="ID Mix (if you want to mix two ID image, please turn this on, otherwise, turn this off)", value=False, ) steps = gr.Slider(label="Steps", value=4, minimum=1, maximum=8, step=1) seed = gr.Slider( label="Seed", value=42, minimum=np.iinfo(np.uint32).min, maximum=np.iinfo(np.uint32).max, step=1 ) n_samples = gr.Slider(label="Num samples", value=2, minimum=1, maximum=4, step=1) scale = gr.Slider( label="CFG, recommend value range [1, 1.5], 1 will be faster ", value=1.2, minimum=1, maximum=1.5, step=0.1, ) with gr.Row(): H = gr.Slider(label="Height", value=1024, minimum=512, maximum=1280, step=64) W = gr.Slider(label="Width", value=768, minimum=512, maximum=1280, step=64) gr.Markdown("## Examples") example_inps = [ [ 'portrait,cinematic,wolf ears,white hair', 'sample_img/sample_img_test24.jpg', 'fidelity', ] ] gr.Examples(examples=example_inps, inputs=[prompt, face_image, mode], label='realistic') example_inps = [ [ 'portrait, impressionist painting, loose brushwork, vibrant color, light and shadow play', 'sample_img/sample_img_test24.jpg', 'fidelity', ] ] gr.Examples(examples=example_inps, inputs=[prompt, face_image, mode], label='painting style') example_inps = [ [ 'portrait, flat papercut style, silhouette, clean cuts, paper, sharp edges, minimalist,color block,man', 'sample_img/sample_img_test1.jpg', 'fidelity', ] ] gr.Examples(examples=example_inps, inputs=[prompt, face_image, mode], label='papercut style') example_inps = [ [ 'woman,cartoon,solo,Popmart Blind Box, Super Mario, 3d', 'sample_img/sample_img_test24.jpg', 'fidelity', ] ] gr.Examples(examples=example_inps, inputs=[prompt, face_image, mode], label='3d style') example_inps = [ [ 'portrait, pixar', 'sample_img/sample_img_test1.jpg', 'fidelity', ] ] gr.Examples(examples=example_inps, inputs=[prompt, face_image, mode], label='portrait style') example_inps = [ [ 'a high quality digital avatar, eating icecream', 'sample_img/image1.png', 'fidelity', ] ] gr.Examples(examples=example_inps, inputs=[prompt, face_image, mode], label='portrait style') example_inps = [ [ 'portrait, the legend of zelda, anime', 'sample_img/image1.png', 'extremely style', ] ] gr.Examples(examples=example_inps, inputs=[prompt, face_image, mode], label='anime style') example_inps = [ [ 'portrait, ironman', 'sample_img/image1.png', 'sample_img/sample_img_test24.jpg', 'fidelity', True, ] ] gr.Examples(examples=example_inps, inputs=[prompt, face_image, supp_image1, mode, id_mix], label='id mix') with gr.Column(): output = gr.Gallery(label='Output', elem_id="gallery") intermediate_output = gr.Gallery(label='DebugImage', elem_id="gallery", visible=False) inps = [ face_image, supp_image1, prompt, neg_prompt, id_scale, mode, id_mix, steps, seed, n_samples, scale, H, W ] submit.click(fn=run, inputs=inps, outputs=[output, intermediate_output]) SAK.launch()