app.py

import gradio as gr
import torch
import spaces
from diffusers import StableDiffusionXLPipeline, DPMSolverMultistepScheduler
from huggingface_hub import hf_hub_download, InferenceClient
import random
import os
import re

# ── Config ────────────────────────────────────────────────────────────────────
HF_TOKEN   = os.environ.get("HF_TOKEN", None)
MODEL_REPO = "John6666/nova-3dcg-xl-illustrious-v40-sdxl"

# Quality tags for Illustrious-based models
IL_POS = "masterpiece, best quality, very aesthetic, absurdres, "
IL_NEG = "worst quality, low quality, bad quality, ugly, "

# ── LLM client ────────────────────────────────────────────────────────────────
llm_client = InferenceClient(
    model="mistralai/Mistral-7B-Instruct-v0.3",
    token=HF_TOKEN,
)

EXPANSION_SYSTEM = """You are an expert Stable Diffusion prompt engineer specialising in 3DCG character art and illustration.

Your job: take a short user description and rewrite it as a detailed, accurate image generation prompt optimised for a 3D CGI character art model (Nova 3DCG XL).

Rules:
- PRESERVE every specific detail — colours, numbers, states, accessories, clothing
- Wrap unique specific details in attention weights e.g. (red scarf:1.4), (one eye closed:1.3)
- Add: character pose, expression, lighting, background atmosphere, material quality, render style
- Add 3DCG-appropriate quality boosters: sharp edges, subsurface scattering, ray tracing, ambient occlusion
- Do NOT add NSFW content
- Do NOT invent things not implied by the user
- Return ONLY the final prompt — no explanation, no preamble, no quotes
- Keep under 130 words
- Use comma-separated tags and phrases"""

def expand_prompt_llm(raw_prompt, style):
    if not raw_prompt.strip():
        return ""
    style_hint = f" The desired style is: {style}." if style != "Auto" else ""
    user_msg = f"Expand this into a detailed 3DCG character art prompt:{style_hint}\n\n{raw_prompt.strip()}"
    try:
        response = llm_client.chat_completion(
            messages=[
                {"role": "system", "content": EXPANSION_SYSTEM},
                {"role": "user",   "content": user_msg},
            ],
            max_tokens=220,
            temperature=0.7,
        )
        expanded = response.choices[0].message.content.strip()
        expanded = expanded.strip('"').strip("'")
        if expanded.lower().startswith("prompt:"):
            expanded = expanded[7:].strip()
        return expanded
    except Exception as e:
        print(f"LLM expansion failed: {e}")
        return raw_prompt.strip()

# ── Load model ────────────────────────────────────────────────────────────────
print(f"Loading Nova 3DCG XL from {MODEL_REPO}...")

pipe = StableDiffusionXLPipeline.from_pretrained(
    MODEL_REPO,
    torch_dtype=torch.float16,
    token=HF_TOKEN,
)
pipe.scheduler = DPMSolverMultistepScheduler.from_config(
    pipe.scheduler.config,
    use_karras_sigmas=True,
)
pipe.enable_attention_slicing()
print("Pipeline ready.")

# ── Style presets ─────────────────────────────────────────────────────────────
STYLES = {
    "Auto":           {"pos": "", "neg": ""},
    "🎮 3DCG Render": {
        "pos": "3DCG render, Pixar style, ray tracing, subsurface scattering, ambient occlusion, sharp edges, studio lighting, ",
        "neg": "flat, 2D, anime flat colour, sketch, ",
    },
    "⚔️ Fantasy":      {
        "pos": "fantasy character, epic armour, magical atmosphere, dramatic lighting, volumetric fog, concept art, artstation, ",
        "neg": "modern, mundane, sci-fi, ",
    },
    "🤖 Sci-Fi":       {
        "pos": "sci-fi character, futuristic suit, neon accents, holographic elements, dark background, cinematic, ",
        "neg": "medieval, fantasy, nature, ",
    },
    "🌸 Stylised":     {
        "pos": "stylised illustration, vibrant colours, soft cel shading, clean lineart, anime-adjacent, ",
        "neg": "photorealistic, gritty, dark, ",
    },
    "🎬 Cinematic":    {
        "pos": "cinematic portrait, dramatic rim lighting, shallow depth of field, film grain, color graded, ",
        "neg": "flat, overexposed, sketch, ",
    },
    "🏙️ Urban":        {
        "pos": "urban streetwear character, city background, neon lights, night scene, realistic clothing, ",
        "neg": "fantasy, medieval, nature, ",
    },
}

# ── LoRAs ─────────────────────────────────────────────────────────────────────
LORAS = {
    "None": None,
    "✋ Better Hands": {
        "repo": "WolfAether21/PONY-DIFFUSION-SDXL-LORA",
        "file": "Perfect Hands v2.safetensors",
        "strength": 0.7,
    },
    "🔍 More Detail": {
        "repo": "WolfAether21/PONY-DIFFUSION-SDXL-LORA",
        "file": "SDXL Detail.safetensors",
        "strength": 0.6,
    },
}

# ── Generation ────────────────────────────────────────────────────────────────
@spaces.GPU(duration=180)
def generate(raw_prompt, negative_prompt, style, lora_name, lora_strength,
             width, height, steps, guidance, seed, randomize, show_expanded):

    if not raw_prompt.strip():
        raise gr.Error("Please enter a prompt.")

    if randomize:
        seed = random.randint(0, 2**32 - 1)
    seed = int(seed)

    # LLM expansion
    expanded = expand_prompt_llm(raw_prompt, style)
    style_data = STYLES.get(style, STYLES["Auto"])
    final_pos  = IL_POS + style_data["pos"] + expanded
    final_neg  = IL_NEG + style_data["neg"] + negative_prompt.strip()

    pipe.to("cuda")

    # LoRA
    lora_loaded = False
    lora_data = LORAS.get(lora_name)
    if lora_data:
        try:
            lp = hf_hub_download(
                repo_id=lora_data["repo"],
                filename=lora_data["file"],
                token=HF_TOKEN,
            )
            pipe.load_lora_weights(lp)
            pipe.fuse_lora(lora_scale=float(lora_strength))
            lora_loaded = True
        except Exception as e:
            print(f"LoRA failed, skipping: {e}")

    generator = torch.Generator(device="cpu").manual_seed(seed)

    result = pipe(
        prompt=final_pos,
        negative_prompt=final_neg,
        width=int(width),
        height=int(height),
        num_inference_steps=int(steps),
        guidance_scale=float(guidance),
        generator=generator,
        clip_skip=1,
    )

    if lora_loaded:
        pipe.unfuse_lora()
        pipe.unload_lora_weights()

    pipe.to("cpu")

    debug = f"**Expanded prompt:**\n\n{final_pos}" if show_expanded else ""
    return result.images[0], seed, debug

# ── CSS ───────────────────────────────────────────────────────────────────────
css = """
* { box-sizing: border-box; margin: 0; padding: 0; }

body, .gradio-container {
    background: #07070e !important;
    font-family: 'Inter', system-ui, -apple-system, sans-serif !important;
    max-width: 500px !important;
    margin: 0 auto !important;
    padding: 8px !important;
}

.topbar {
    display: flex;
    align-items: center;
    justify-content: space-between;
    padding: 10px 2px 14px;
}
.topbar-title {
    color: #e8e0ff;
    font-size: 0.95em;
    font-weight: 800;
}
.gpu-pill {
    background: #1aff7a18;
    border: 1px solid #1aff7a44;
    color: #1aff7a;
    font-size: 0.6em;
    font-weight: 800;
    padding: 4px 12px;
    border-radius: 20px;
    letter-spacing: 1.5px;
    text-transform: uppercase;
}

.img-out {
    background: #0d0d1a;
    border: 1px solid #16162a;
    border-radius: 20px;
    overflow: hidden;
    margin-bottom: 8px;
    min-height: 380px;
    display: flex;
    align-items: center;
    justify-content: center;
}
.img-out img {
    width: 100% !important;
    border-radius: 20px;
    display: block;
}

.seed-pill input[type=number] {
    background: transparent !important;
    border: none !important;
    color: #2e2848 !important;
    font-size: 0.7em !important;
    text-align: center !important;
    padding: 2px !important;
    width: 100% !important;
}

.card {
    background: #0d0d1a;
    border: 1px solid #16162a;
    border-radius: 14px;
    padding: 14px;
    margin-bottom: 8px;
}
.card-label {
    color: #3d3060;
    font-size: 0.62em;
    font-weight: 800;
    text-transform: uppercase;
    letter-spacing: 2px;
    margin-bottom: 8px;
}

textarea {
    background: transparent !important;
    border: none !important;
    color: #c8b8f0 !important;
    font-size: 15px !important;
    line-height: 1.6 !important;
    padding: 0 !important;
    resize: none !important;
    box-shadow: none !important;
    width: 100% !important;
    outline: none !important;
}
textarea::placeholder { color: #252038 !important; }
textarea:focus {
    outline: none !important;
    box-shadow: none !important;
    border: none !important;
}

.style-wrap .gr-radio {
    display: flex !important;
    flex-wrap: wrap !important;
    gap: 6px !important;
}
.style-wrap label {
    background: #0d0d1a !important;
    border: 1px solid #1a1a2e !important;
    border-radius: 30px !important;
    color: #4a3a6a !important;
    font-size: 0.75em !important;
    font-weight: 600 !important;
    padding: 6px 14px !important;
    cursor: pointer !important;
    transition: all 0.15s ease !important;
    white-space: nowrap !important;
}
.style-wrap label:has(input:checked) {
    background: #18083a !important;
    border-color: #7744ee !important;
    color: #bb99ff !important;
    box-shadow: 0 0 10px #7744ee33 !important;
}
.style-wrap input[type=radio] { display: none !important; }

.gradio-accordion {
    background: #0d0d1a !important;
    border: 1px solid #16162a !important;
    border-radius: 14px !important;
    margin-bottom: 8px !important;
    overflow: hidden !important;
}
.gradio-accordion .label-wrap button {
    color: #4a3a6a !important;
    font-size: 0.72em !important;
    font-weight: 700 !important;
    text-transform: uppercase !important;
    letter-spacing: 1.5px !important;
    padding: 12px 16px !important;
}

.gradio-slider {
    background: transparent !important;
    border: none !important;
    padding: 4px 0 10px !important;
}
input[type=range] {
    accent-color: #6633bb !important;
    width: 100% !important;
}

input[type=number] {
    background: #0a0a14 !important;
    border: 1px solid #18182a !important;
    border-radius: 10px !important;
    color: #9977cc !important;
    font-size: 13px !important;
    padding: 8px 10px !important;
}

input[type=checkbox] { accent-color: #6633bb !important; }
.gradio-checkbox label span {
    color: #4a3a6a !important;
    font-size: 0.75em !important;
    font-weight: 600 !important;
}

.gradio-dropdown {
    background: #0a0a14 !important;
    border: 1px solid #18182a !important;
    border-radius: 10px !important;
}

label > span:first-child {
    color: #3a2d55 !important;
    font-size: 0.7em !important;
    font-weight: 700 !important;
    text-transform: uppercase !important;
    letter-spacing: 1px !important;
}

.debug-box {
    background: #080814;
    border: 1px solid #111122;
    border-radius: 10px;
    padding: 10px 12px;
    color: #443366;
    font-size: 0.7em;
    line-height: 1.7;
    font-family: monospace;
    word-break: break-word;
    margin-bottom: 8px;
    min-height: 10px;
}

.gen-btn button {
    background: linear-gradient(135deg, #4a1aaa 0%, #2d0e77 100%) !important;
    border: 1px solid #6633cc !important;
    border-radius: 14px !important;
    color: #fff !important;
    font-size: 0.88em !important;
    font-weight: 900 !important;
    padding: 17px !important;
    width: 100% !important;
    letter-spacing: 2px !important;
    text-transform: uppercase !important;
    box-shadow: 0 4px 24px #4a1aaa55 !important;
    transition: all 0.15s ease !important;
    margin-top: 6px !important;
}
.gen-btn button:hover {
    box-shadow: 0 6px 32px #4a1aaa99 !important;
    transform: translateY(-1px) !important;
}
.gen-btn button:active {
    transform: scale(0.98) !important;
    box-shadow: 0 2px 12px #4a1aaa33 !important;
}

footer, .built-with { display: none !important; }
"""

# ── UI ────────────────────────────────────────────────────────────────────────
with gr.Blocks(css=css, title="ImageGen") as demo:

    gr.HTML("""
    <div class="topbar">
        <span class="topbar-title">Nova 3DCG XL</span>
        <span class="gpu-pill">⚡ ZeroGPU</span>
    </div>
    """)

    output_image = gr.Image(
        show_label=False, type="pil",
        height=460, elem_classes="img-out",
    )
    used_seed = gr.Number(
        label="seed", interactive=False,
        elem_classes="seed-pill",
    )

    gr.HTML('<div class="card"><div class="card-label">✦ Prompt — describe your character</div>')
    prompt = gr.Textbox(
        show_label=False,
        placeholder="warrior woman in red armour, glowing sword, forest background...",
        lines=3,
    )
    gr.HTML('</div>')

    gr.HTML('<div class="card-label" style="padding:4px 2px 8px;color:#3d3060;font-size:0.62em;font-weight:800;text-transform:uppercase;letter-spacing:2px;">Style</div>')
    style = gr.Radio(
        choices=list(STYLES.keys()),
        value="Auto",
        show_label=False,
        elem_classes="style-wrap",
    )

    generate_btn = gr.Button(
        "Generate ✦", variant="primary",
        size="lg", elem_classes="gen-btn",
    )

    expanded_out = gr.Markdown(
        value="",
        elem_classes="debug-box",
    )

    with gr.Accordion("⚙️  Settings", open=False):
        gr.HTML('<div style="height:6px"></div>')

        negative_prompt = gr.Textbox(
            label="Negative Prompt",
            value=(
                "worst quality, low quality, bad anatomy, bad hands, "
                "extra limbs, missing limbs, watermark, signature, "
                "blurry, deformed, ugly, text"
            ),
            lines=2,
        )

        with gr.Row():
            width  = gr.Slider(512, 1024, value=832,  step=64, label="Width")
            height = gr.Slider(512, 1216, value=1216, step=64, label="Height")

        steps    = gr.Slider(20, 60,    value=30,  step=1,   label="Steps")
        guidance = gr.Slider(1.0, 10.0, value=6.0, step=0.5, label="CFG Scale")

        with gr.Row():
            seed = gr.Number(
                label="Seed", value=42, precision=0,
                minimum=0, maximum=2**32-1, scale=3,
            )
            randomize = gr.Checkbox(label="Random seed", value=True, scale=1)

        show_expanded = gr.Checkbox(
            label="Show expanded prompt",
            value=True,
        )

    with gr.Accordion("🎨  LoRA", open=False):
        gr.HTML('<div style="height:6px"></div>')
        lora_name     = gr.Dropdown(choices=list(LORAS.keys()), value="None", label="LoRA")
        lora_strength = gr.Slider(0.1, 1.0, value=0.7, step=0.05, label="Strength")

    generate_btn.click(
        fn=generate,
        inputs=[
            prompt, negative_prompt, style, lora_name, lora_strength,
            width, height, steps, guidance, seed, randomize, show_expanded,
        ],
        outputs=[output_image, used_seed, expanded_out],
    )

demo.launch()