bleckhert's picture
Upload app.py with huggingface_hub
aea9dc2 verified
#!/usr/bin/env python3
import gradio as gr
import numpy as np
from PIL import Image, ImageDraw, ImageFont
import cv2
import os
import json
import hashlib
import time
import random
from datetime import datetime
from pathlib import Path
import warnings
warnings.filterwarnings("ignore")
# All heavy imports happen inside button handlers
print("App loaded successfully")
class SafetyFramework:
BLOCKED = ["child","minor","underage","kid","children","non-consensual","revenge",
"hidden camera","spy","torture","gore","snuff","beheading","execution",
"terrorist","bomb making","how to kill"]
WARN = ["real person","celebrity","famous","actor","actress","public figure","politician","named individual"]
def check_prompt(self, text):
if not text: return True, "ok", ""
text_lower = text.lower()
for kw in self.BLOCKED:
if kw in text_lower:
return False, "blocked", f"Blocked term: '{kw}'"
wf = [k for k in self.WARN if k in text_lower]
if wf:
return True, "warning", f"Warning: references real individuals ({', '.join(wf)})"
return True, "ok", ""
def watermark(self, image, meta=None):
if image is None: return image
img = image.copy().convert("RGBA")
w, h = img.size
overlay = Image.new("RGBA", img.size, (0,0,0,0))
draw = ImageDraw.Draw(overlay)
try: font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", max(10, h//80))
except: font = ImageFont.load_default()
wm = "AI GENERATED - Fictional Content"
ts = datetime.utcnow().strftime("%Y-%m-%d %H:%M UTC")
try:
bx = draw.textbbox((0,0), wm, font=font)
tw, th = bx[2]-bx[0], bx[3]-bx[1]
except:
tw, th = len(wm)*6, 12
x, y = w-tw-15, h-th-15
draw.rectangle([x-5, y-5, x+tw+5, y+th+5], fill=(0,0,0,80))
draw.text((x,y), wm, fill=(255,255,255,180), font=font)
draw.rectangle([x-5, y-th-7, x+len(ts)*6+5, y-5], fill=(0,0,0,80))
draw.text((x, y-th-5), ts, fill=(200,200,200,150), font=font)
return Image.alpha_composite(img, overlay).convert("RGB")
SAFETY = SafetyFramework()
def placeholder(prompt, w, h, label="", sub=""):
img = Image.new("RGB", (int(w), int(h)), (30,30,40))
d = ImageDraw.Draw(img)
try:
fl = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 24)
fs = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 14)
except:
fl = fs = ImageFont.load_default()
t = "AI Creative Suite"
d.text(((int(w)-len(t)*14)//2, 20), t, fill=(200,200,255), font=fl)
p = prompt[:60] + "..." if len(prompt)>60 else prompt
d.text(((int(w)-len(p)*8)//2, int(h)//2-30), p, fill=(180,180,200), font=fs)
if sub: d.text(((int(w)-len(sub)*8)//2, int(h)//2), sub, fill=(150,150,180), font=fs)
if label: d.text(((int(w)-len(label)*8)//2, int(h)//2+30), label, fill=(150,150,180), font=fs)
return SAFETY.watermark(img)
class ImageGen:
def __init__(self):
self._pipe = None
def _init(self):
if self._pipe is not None: return True
try:
import torch
from diffusers import DiffusionPipeline
dev = "cuda" if torch.cuda.is_available() else "cpu"
self._pipe = DiffusionPipeline.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0",
torch_dtype=torch.float16 if dev=="cuda" else torch.float32,
use_safetensors=True, variant="fp16" if dev=="cuda" else None)
if dev=="cuda":
self._pipe = self._pipe.to(dev)
self._pipe.enable_model_cpu_offload()
self._dev = dev
return True
except Exception as e:
self._err = str(e)
return False
def generate(self, prompt, neg, width, height, steps, guidance, seed, num):
ok, lvl, msg = SAFETY.check_prompt(prompt)
if not ok:
return [placeholder(prompt, width, height, "BLOCKED", msg)]*int(num), msg
if not self._init():
return [placeholder(prompt, width, height, "LOADING", self._err)]*int(num), f"Model not ready: {self._err}"
import torch
s = int(seed) if int(seed)!=-1 else random.randint(0, 2**32)
gen = torch.Generator(device=self._dev).manual_seed(s)
out = []
for i in range(int(num)):
g = gen.manual_seed(s+i)
try:
r = self._pipe(prompt=prompt, negative_prompt=neg, width=int(width), height=int(height),
num_inference_steps=int(steps), guidance_scale=float(guidance), generator=g).images[0]
out.append(SAFETY.watermark(r, {"seed": s+i}))
except Exception as e:
out.append(placeholder(prompt, width, height, "ERROR", str(e)[:50]))
return out, msg if lvl=="warning" else ""
class VideoGen:
def __init__(self): self._pipe = None
def _init(self):
if self._pipe is not None: return True
try:
import torch
from diffusers import CogVideoXPipeline
self._pipe = CogVideoXPipeline.from_pretrained("THUDM/CogVideoX-2b",
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32)
if torch.cuda.is_available(): self._pipe.enable_model_cpu_offload()
return True
except Exception as e:
self._err = str(e)
return False
def generate(self, prompt, num_frames, fps, seed):
ok, lvl, msg = SAFETY.check_prompt(prompt)
if not ok: return self._fallback(prompt, num_frames, fps, "BLOCKED"), msg
if self._init():
try:
import torch
s = int(seed) if int(seed)!=-1 else random.randint(0, 2**32)
gen = torch.Generator(device="cuda" if torch.cuda.is_available() else "cpu").manual_seed(s)
v = self._pipe(prompt=prompt, num_frames=int(num_frames), num_inference_steps=50,
guidance_scale=6.0, generator=gen).frames[0]
path = f"/tmp/video_{int(time.time())}.mp4"
from diffusers.utils import export_to_video
export_to_video(v, path, fps=int(fps))
return path, msg if lvl=="warning" else ""
except Exception as e:
return self._fallback(prompt, num_frames, fps, str(e)[:50]), str(e)
return self._fallback(prompt, num_frames, fps, self._err), self._err
def _fallback(self, prompt, nf, fps, status=""):
path = f"/tmp/vid_{int(time.time())}.mp4"
frames = []
w, h = 512, 512
try:
f = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 20)
fs = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 14)
except:
f = fs = ImageFont.load_default()
for i in range(int(nf)):
img = Image.new("RGB", (w,h), (20,20,30))
d = ImageDraw.Draw(img)
off = int(15*np.sin(i*2*3.14159/max(nf,1)))
d.text((w//2-100+off, 30), "AI Creative Suite - Video", fill=(200,200,255), font=f)
p = prompt[:50]+"..." if len(prompt)>50 else prompt
d.text((w//2-len(p)*4, h//2-20), p, fill=(180,180,200), font=fs)
if status: d.text((20, h//2+10), f"Status: {status}", fill=(255,100,100), font=fs)
bw = 300; bx = (w-bw)//2; by = h//2+40
pr = (i+1)/max(nf,1); fw = int(bw*pr)
d.rectangle([bx,by,bx+bw,by+20], outline=(100,100,150), width=2)
d.rectangle([bx+2,by+2,bx+fw-2,by+18], fill=(100,150,255))
frames.append(np.array(img))
try:
import imageio
imageio.mimsave(path, frames, fps=int(fps))
except:
img = Image.new("RGB",(w,h),(20,20,30)); ImageDraw.Draw(img).text((50,200),"Video error",fill=(200,200,255)); img.save(path)
return path
class AudioGen:
def __init__(self):
self._music = None; self._sfx = None
def _init_music(self, size):
try:
from transformers import AutoProcessor, MusicgenForConditionalGeneration
p = AutoProcessor.from_pretrained(f"facebook/musicgen-{size}")
m = MusicgenForConditionalGeneration.from_pretrained(f"facebook/musicgen-{size}")
self._music = (p, m); return True
except Exception as e: self._music_err = str(e); return False
def _init_sfx(self):
try:
from transformers import AutoProcessor, AutoModelForTextToWaveform
p = AutoProcessor.from_pretrained("facebook/audiogen-medium")
m = AutoModelForTextToWaveform.from_pretrained("facebook/audiogen-medium")
self._sfx = (p, m); return True
except Exception as e: self._sfx_err = str(e); return False
def generate_music(self, prompt, duration, seed, size):
ok, lvl, msg = SAFETY.check_prompt(prompt)
if not ok: return self._placeholder(float(duration), "BLOCKED"), msg
if not self._music and not self._init_music(size):
return self._placeholder(float(duration), self._music_err), self._music_err
try:
import torch
p, m = self._music
inputs = p(text=[prompt], padding=True, return_tensors="pt")
mt = min(int(float(duration)*50), 1500)
av = m.generate(**inputs, max_new_tokens=mt, do_sample=True, guidance_scale=3.0)
path = f"/tmp/mus_{int(time.time())}.wav"
import scipy.io.wavfile
scipy.io.wavfile.write(path, rate=m.config.audio_encoder.sampling_rate, data=av[0,0].cpu().numpy())
return path, msg if lvl=="warning" else ""
except Exception as e: return self._placeholder(float(duration), str(e)[:60]), str(e)
def generate_sfx(self, prompt, duration, seed):
ok, lvl, msg = SAFETY.check_prompt(prompt)
if not ok: return self._placeholder(float(duration), "BLOCKED"), msg
if not self._sfx and not self._init_sfx():
return self._placeholder(float(duration), self._sfx_err), self._sfx_err
try:
p, m = self._sfx
inputs = p(text=[prompt], return_tensors="pt")
mt = min(int(float(duration)*50), 1000)
av = m.generate(**inputs, max_new_tokens=mt, do_sample=True)
path = f"/tmp/sfx_{int(time.time())}.wav"
import scipy.io.wavfile
scipy.io.wavfile.write(path, rate=16000, data=av[0,0].cpu().numpy())
return path, msg if lvl=="warning" else ""
except Exception as e: return self._placeholder(float(duration), str(e)[:60]), str(e)
def _placeholder(self, dur, label=""):
sr = 16000; s = np.zeros(int(sr*dur), np.float32)
t = np.linspace(0,dur,len(s))
s += 0.1*np.sin(2*3.14159*440*t)*(t<0.5)
s += 0.05*np.sin(2*3.14159*880*t)*((t>0.5)&(t<1.0))
path = f"/tmp/ph_{int(time.time())}.wav"
try:
import scipy.io.wavfile
scipy.io.wavfile.write(path, rate=sr, data=s)
except:
with open(path, "wb") as f: pass
return path
class Enhancer:
def upscale(self, image, scale):
if image is None: return None
w, h = image.size
ns = (int(w*float(scale)), int(h*float(scale)))
return SAFETY.watermark(image.resize(ns, Image.Resampling.LANCZOS))
def skin(self, image, strength):
if image is None: return None
img = np.array(image).astype(np.float32)
sm = cv2.bilateralFilter(img.astype(np.uint8), 9, 75, 75)
en = img*(1-float(strength)) + sm.astype(np.float32)*float(strength)
k = np.array([[-1,-1,-1],[-1,9,-1],[-1,-1,-1]])*0.3 + np.array([[0,0,0],[0,1,0],[0,0,0]])*0.7
en = cv2.filter2D(en.astype(np.uint8), -1, k)
return SAFETY.watermark(Image.fromarray(en))
class PlotGen:
def __init__(self):
self.tmpl = {
"romance": {"scenes":["Meeting","Connection","Conflict","Resolution","Intimacy"],
"beats":["first glance","shared secret","external obstacle","emotional breakthrough","physical closeness"]},
"adventure": {"scenes":["Departure","Trials","Discovery","Confrontation","Return"],
"beats":["call to action","overcoming fear","hidden truth","final battle","changed perspective"]},
"mystery": {"scenes":["Incident","Investigation","Twist","Confrontation","Revelation"],
"beats":["unexplained event","clue gathering","false lead","accusation","truth uncovered"]},
"fantasy": {"scenes":["Ordinary World","Crossing","Allies","Ordeal","Mastery"],
"beats":["mundane life","portal opens","unlikely friendship","greatest fear","new power"]},
"thriller": {"scenes":["Calm","Disturbance","Escalation","Crisis","Aftermath"],
"beats":["peaceful moment","unusual detail","stakes rise","point of no return","new normal"]},
"sci-fi": {"scenes":["Present","Anomaly","Exploration","Revelation","Transformation"],
"beats":["technological world","strange signal","unknown territory","alien truth","human evolution"]}
}
self.emos = {"passionate":["intense gaze","trembling touch","racing heartbeats","heated whisper","burning desire"],
"tense":["clenched jaw","narrowed eyes","heavy silence","shallow breathing","coiled energy"],
"joyful":["bright laughter","warm embrace","sparkling eyes","relaxed posture","genuine smile"],
"mysterious":["shadowed face","half-smile","glance over shoulder","unspoken knowledge","concealed intention"],
"dark":["haunted expression","clenched fists","distant stare","sharp movements","controlled rage"]}
def generate(self, genre, theme, tone, num_scenes, setting, chars):
t = self.tmpl.get(genre, self.tmpl["romance"])
e = self.emos.get(tone, self.emos["passionate"])
n = min(int(num_scenes), len(t["scenes"]))
out = f"# {genre.upper()} PLOT: {theme or 'Untitled'}\n\n"
out += f"**Tone:** {tone} | **Setting:** {setting or 'Various'} | **Characters:** {chars or '2'}\n\n---\n\n"
for i in range(n):
sn = t["scenes"][i]; bt = t["beats"][i] if i<len(t["beats"]) else "development"
em = random.choice(e)
out += f"## Scene {i+1}: {sn}\n\n"
out += f"**Theme:** {theme or 'journey'} | **Beat:** {bt} | **Emotion:** {em}\n\n"
out += f"**Description:** {self._desc(sn, bt, tone, em, theme, setting, chars)}\n\n"
out += f"**Actions:** {self._act(sn)}\n\n"
out += f"**Visual Prompt:** {self._vis(sn, tone, theme, setting, em)}\n\n"
out += f"**Audio:** {self._aud(sn, tone, em)}\n\n"
out += f"**Duration:** {random.choice([3,5,8,10])}s\n\n---\n\n"
out += "## Production Notes\n\n"
out += f"- Runtime: ~{sum([random.choice([3,5,8,10]) for _ in range(n)])}s\n"
out += f"- Color: {random.choice(['warm golds','cool blues','high contrast','muted earth','neon accents'])}\n"
out += f"- Camera: {random.choice(['steady wides','intimate close-ups','handheld doc','sweeping drones'])}\n"
out += f"- Pacing: {random.choice(['slow contemplative','fast cuts','builds to climax','rhythmic'])}\n"
return out
def _desc(self, s, b, t, e, th, st, c):
defs = {k: [f"The {c or 'two'} in {st or 'the scene'}, {e}, {t} atmosphere."] for k,v in self.tmpl["romance"]["scenes"].items() if False}
# default
return f"The {t} scene unfolds in {st or 'the moment'}, carrying {th or 'the story'} with {e}."
def _act(self, s):
acts = {"Meeting":"exchange glances, approach cautiously, initiate conversation",
"Connection":"lean in closer, share a secret, laugh together, touch hands",
"Conflict":"turn away sharply, raise voice, clench fists, walk out",
"Resolution":"embrace tightly, whisper apologies, hold each other",
"Intimacy":"caress gently, move together, express love verbally"}
return acts.get(s, "interact meaningfully, exchange looks")
def _vis(self, s, t, th, st, e):
v = {"Meeting":f"cinematic shot, characters first meeting in {st or 'elegant location'}, {t} atmosphere, {e}, dramatic lighting",
"Connection":f"intimate scene, bonding, {t} mood, {st or 'soft lighting'}, {e}, emotional depth",
"Conflict":f"dramatic confrontation, {t} tension, {st or 'dramatic location'}, {e}, high stakes",
"Resolution":f"emotional resolution, {t} mood, {st or 'golden hour'}, {e}, beautiful cinematography",
"Intimacy":f"intimate moment, {t} atmosphere, {st or 'warm space'}, {e}, tender connection, soft focus"}
return v.get(s, f"cinematic scene, {t} mood, {st or 'dramatic lighting'}, {e}")
def _aud(self, s, t, e):
a = {"Meeting":f"soft ambient building to strings, {t} undertones, {e} harmonies",
"Connection":f"gentle piano with emotional resonance, {t} harmonies, {e} textures",
"Conflict":f"rising tension with percussion, {t} dissonance, {e} crescendo",
"Resolution":f"orchestral triumph with brass, {t} resolution, {e} release",
"Intimacy":f"soft sensual ambient pads, {t} warmth, {e} intimacy"}
return a.get(s, f"{t} ambient soundtrack with {e} textures")
class FilmEd:
def __init__(self): self._proj = {}
def create(self, name):
if not name: return "Provide project name"
self._proj[name] = {"scenes": [], "created": datetime.utcnow().isoformat()}
return f"Project '{name}' created"
def add(self, proj, img, dur, aud):
if proj not in self._proj: return f"Create '{proj}' first"
if img is None: return "Provide image"
sid = len(self._proj[proj]["scenes"])
self._proj[proj]["scenes"].append({"id":sid, "image":img, "duration":float(dur), "audio":aud})
return f"Scene {sid} added. Total: {sid+1}"
def render(self, proj, fps, trans):
if proj not in self._proj: return None, f"Create '{proj}' first"
sc = self._proj[proj]["scenes"]
if not sc: return None, "No scenes"
frames = []
for s in sc:
img = s["image"].copy().convert("RGB").resize((512,512), Image.Resampling.LANCZOS)
ov = Image.new("RGBA", img.size, (0,0,0,0))
d = ImageDraw.Draw(ov)
try: fnt = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 16)
except: fnt = ImageFont.load_default()
d.rectangle([5,5,200,30], fill=(0,0,0,150))
d.text((10,8), f"Scene {s['id']} | {s['duration']}s", fill=(255,255,255), font=fnt)
if trans != "none":
tt = f"Transition: {trans}"
d.rectangle([512-len(tt)*9-20,5,507,30], fill=(0,0,0,150))
d.text((512-len(tt)*9-15,8), tt, fill=(200,200,255), font=fnt)
d.rectangle([5,487,120,507], fill=(0,0,0,150))
d.text((10,490), "AI GENERATED", fill=(255,255,255), font=fnt)
img = Image.alpha_composite(img.convert("RGBA"), ov).convert("RGB")
for _ in range(int(s["duration"]*int(fps))): frames.append(np.array(img))
path = f"/tmp/film_{proj}_{int(time.time())}.mp4"
try:
import imageio
imageio.mimsave(path, frames, fps=int(fps))
except Exception as e:
img = Image.new("RGB",(512,512),(20,20,30)); ImageDraw.Draw(img).text((50,200),f"Film: {proj}",fill=(200,200,255)); ImageDraw.Draw(img).text((50,250),f"Err: {str(e)[:40]}",fill=(150,150,180)); img.save(path)
return path, f"Film '{proj}' rendered: {len(sc)} scenes at {fps} fps"
class PoseEd:
def extract(self, image):
if image is None: return None
try:
from controlnet_aux import OpenposeDetector
return OpenposeDetector.from_pretrained("lllyasviel/ControlNet")(image)
except:
e = cv2.Canny(np.array(image.convert("L")), 100, 200)
return Image.fromarray(cv2.cvtColor(e, cv2.COLOR_GRAY2RGB))
def views(self, image, prompt, n):
if image is None: return [], "No image"
ok, lvl, msg = SAFETY.check_prompt(prompt)
if not ok: return [placeholder(prompt, image.size[0], image.size[1], "BLOCKED")]*int(n), msg
v = ["front view","side profile","three-quarter view","back view","high angle","low angle","close-up","wide shot"]
r = []
for a in v[:int(n)]:
img = image.copy().convert("RGB")
ov = Image.new("RGBA", img.size, (0,0,0,0))
d = ImageDraw.Draw(ov)
try: fnt = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 32)
except: fnt = ImageFont.load_default()
d.rectangle([img.size[0]//2-60,20,img.size[0]//2+60,60], fill=(0,0,0,180))
d.text((img.size[0]//2-50,25), a.upper(), fill=(255,200,100), font=fnt)
r.append(SAFETY.watermark(Image.alpha_composite(img.convert("RGBA"), ov).convert("RGB"), {"view":a}))
return r, msg if lvl=="warning" else ""
class VarGen:
def gen(self, image, prompt, n, seed):
if image is None: return [], "No image"
ok, lvl, msg = SAFETY.check_prompt(prompt)
if not ok: return [placeholder(prompt, image.size[0], image.size[1], "BLOCKED")]*int(n), msg
styles = ["cinematic","oil painting","digital art","watercolor","film noir","neon glow","golden hour","studio lighting"]
r = []
for i, s in enumerate(styles[:int(n)]):
img = image.copy().convert("RGB")
if "cinematic" in s or "film" in s: img = img.filter(ImageFilter.UnsharpMask(radius=2, percent=150, threshold=3))
elif "oil" in s: img = img.filter(ImageFilter.MedianFilter(size=5))
elif "watercolor" in s: img = img.filter(ImageFilter.SMOOTH_MORE)
elif "neon" in s:
try:
from PIL import ImageEnhance
img = ImageEnhance.Color(img).enhance(2.0)
except: pass
elif "golden" in s:
img = Image.blend(img, Image.new("RGB", img.size, (255,200,100)), 0.15)
ov = Image.new("RGBA", img.size, (0,0,0,0))
d = ImageDraw.Draw(ov)
try: fnt = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", 28)
except: fnt = ImageFont.load_default()
l = f"Var {i+1}: {s}"
d.rectangle([10,img.size[1]-50,10+len(l)*14+20,img.size[1]-10], fill=(0,0,0,180))
d.text((20,img.size[1]-45), l, fill=(200,255,200), font=fnt)
r.append(SAFETY.watermark(Image.alpha_composite(img.convert("RGBA"), ov).convert("RGB"), {"style":s}))
return r, msg if lvl=="warning" else ""
def build_ui():
with gr.Blocks(title="AI Creative Production Suite", css="""
.safety-notice { background:#fff3cd; border-left:4px solid #ffc107; padding:12px; margin:10px 0; border-radius:4px; }
.info-box { background:#e7f3ff; border-left:4px solid #2196F3; padding:12px; margin:10px 0; border-radius:4px; }
""") as demo:
gr.Markdown("""
# 🎬 AI Creative Production Suite
<div class="safety-notice">
⚠️ <strong>Safety:</strong> Watermarked outputs. Fictional characters only. No face-swap or voice cloning.
</div>
""")
with gr.Tab("Image Generator"):
with gr.Row():
with gr.Column(scale=2):
gp = gr.Textbox(label="Prompt", placeholder="cinematic portrait of a fantasy warrior, dramatic lighting", lines=3)
gn = gr.Textbox(label="Negative Prompt", value="blurry, low quality, distorted, deformed, extra limbs", lines=2)
with gr.Row():
gw = gr.Slider(512, 1536, 1024, 64, label="Width")
gh = gr.Slider(512, 1536, 1024, 64, label="Height")
with gr.Row():
gs = gr.Slider(10, 100, 30, 1, label="Steps")
gg = gr.Slider(1, 20, 7.5, 0.5, label="Guidance")
with gr.Row():
gd = gr.Number(-1, label="Seed (-1=random)", precision=0)
gn_img = gr.Slider(1, 4, 1, 1, label="Num Images")
gb = gr.Button("Generate", variant="primary")
gw_txt = gr.Textbox(label="Safety Check", interactive=False)
with gr.Column(scale=3):
go = gr.Gallery(label="Generated Images", columns=2, rows=2)
with gr.Tab("Image Enhancer"):
with gr.Row():
with gr.Column():
ei = gr.Image(label="Input Image", type="pil")
es = gr.Slider(1, 4, 2, 0.5, label="Upscale Factor")
esk = gr.Slider(0, 1, 0.5, 0.1, label="Skin Texture Strength")
with gr.Row():
ebu = gr.Button("Upscale")
ebs = gr.Button("Enhance Texture")
with gr.Column():
eo = gr.Image(label="Enhanced Result")
with gr.Tab("Pose Editor"):
with gr.Row():
with gr.Column():
pi = gr.Image(label="Character Reference", type="pil")
pp = gr.Textbox(label="Character Description", placeholder="female warrior with silver armor", lines=2)
pn = gr.Slider(1, 8, 4, 1, label="Number of Views")
with gr.Row():
pbx = gr.Button("Extract Pose")
pbg = gr.Button("Generate Views", variant="primary")
pw = gr.Textbox(label="Safety Check", interactive=False)
with gr.Column():
po = gr.Gallery(label="Generated Views", columns=2, rows=2)
with gr.Tab("Image Variations"):
with gr.Row():
with gr.Column():
vi = gr.Image(label="Reference Image", type="pil")
vp = gr.Textbox(label="Base Prompt", placeholder="portrait of a character", lines=2)
vn = gr.Slider(1, 8, 4, 1, label="Variations")
vs = gr.Number(-1, label="Seed", precision=0)
vb = gr.Button("Generate Variations", variant="primary")
vw = gr.Textbox(label="Safety Check", interactive=False)
with gr.Column():
vo = gr.Gallery(label="Style Variations", columns=2, rows=2)
with gr.Tab("Video Generator"):
with gr.Row():
with gr.Column():
vdp = gr.Textbox(label="Video Prompt", placeholder="slow motion ocean waves, golden hour, cinematic", lines=3)
with gr.Row():
vdf = gr.Slider(16, 49, 25, 1, label="Frames")
vdfps = gr.Slider(4, 30, 8, 1, label="FPS")
vds = gr.Number(-1, label="Seed", precision=0)
vdb = gr.Button("Generate Video", variant="primary")
vdw = gr.Textbox(label="Safety Check", interactive=False)
with gr.Column():
vdo = gr.Video(label="Generated Video")
with gr.Tab("Audio Generator"):
with gr.Row():
with gr.Column():
gr.Markdown("#### Music")
amp = gr.Textbox(label="Music Prompt", placeholder="romantic orchestral music, soft piano", lines=2)
with gr.Row():
amd = gr.Slider(5, 60, 10, 5, label="Duration (s)")
amsz = gr.Radio(["small","medium","large"], value="small", label="Size")
ams = gr.Number(-1, label="Seed", precision=0)
amb = gr.Button("Generate Music", variant="primary")
amw = gr.Textbox(label="Safety Check", interactive=False)
with gr.Column():
amo = gr.Audio(label="Generated Music", type="filepath")
with gr.Row():
with gr.Column():
gr.Markdown("#### Sound Effects")
asp = gr.Textbox(label="SFX Prompt", placeholder="rain on window, footsteps on gravel", lines=2)
with gr.Row():
asd = gr.Slider(1, 30, 5, 1, label="Duration (s)")
ass = gr.Number(-1, label="Seed", precision=0)
asb = gr.Button("Generate SFX")
asw = gr.Textbox(label="Safety Check", interactive=False)
with gr.Column():
aso = gr.Audio(label="Generated SFX", type="filepath")
gr.Markdown("<div class='safety-notice'>⚠️ Instrumental music and environmental sounds only. No voice cloning.</div>")
with gr.Tab("Plot Generator"):
with gr.Row():
with gr.Column():
plg = gr.Dropdown(["romance","adventure","mystery","fantasy","thriller","sci-fi"], value="romance", label="Genre")
plt = gr.Textbox(label="Theme/Topic", placeholder="forbidden love, time travel")
plto = gr.Dropdown(["passionate","tense","joyful","mysterious","dark"], value="passionate", label="Tone")
with gr.Row():
pln = gr.Slider(3, 10, 5, 1, label="Scenes")
plc = gr.Textbox(label="Characters", value="2")
pls = gr.Textbox(label="Setting", placeholder="Victorian mansion, futuristic city")
plb = gr.Button("Generate Plot", variant="primary")
with gr.Column(scale=2):
plo = gr.Textbox(label="Generated Plot", lines=40)
with gr.Tab("Film Editor"):
with gr.Row():
with gr.Column():
gr.Markdown("#### Project")
fpn = gr.Textbox(label="Project Name", value="my_film")
with gr.Row():
fpc = gr.Button("Create Project")
fps = gr.Textbox(label="Status", interactive=False)
gr.Markdown("---")
gr.Markdown("#### Add Scene")
fsi = gr.Image(label="Scene Image", type="pil")
with gr.Row():
fsd = gr.Slider(1, 30, 5, 1, label="Duration (s)")
fsa = gr.Audio(label="Audio (optional)", type="filepath")
fpa = gr.Button("Add Scene")
gr.Markdown("---")
gr.Markdown("#### Render")
with gr.Row():
fpf = gr.Slider(12, 60, 24, 1, label="FPS")
fpt = gr.Dropdown(["none","fade","cut","wipe"], value="fade", label="Transition")
fpr = gr.Button("Render Film", variant="primary")
with gr.Column():
fpo = gr.Video(label="Rendered Film")
fprs = gr.Textbox(label="Render Status", interactive=False)
with gr.Tab("Prompt Helper"):
with gr.Row():
with gr.Column():
phb = gr.Textbox(label="Basic Prompt", placeholder="portrait of a warrior", lines=2)
with gr.Row():
phst = gr.Dropdown(["cinematic","oil painting","digital art","anime","photorealistic","fantasy art","watercolor","film noir"], value="cinematic", label="Style")
phl = gr.Dropdown(["golden hour","dramatic lighting","soft natural","neon","moonlight","studio lighting","volumetric fog"], value="dramatic lighting", label="Lighting")
phq = gr.Dropdown(["masterpiece","highly detailed","8k resolution","concept art","trending on artstation","award winning"], value="masterpiece, highly detailed", label="Quality")
phc = gr.Dropdown(["close-up portrait","wide shot","medium shot","extreme close-up","overhead shot","low angle"], value="close-up portrait", label="Camera")
phbtn = gr.Button("Enhance Prompt", variant="primary")
with gr.Column():
pho = gr.Textbox(label="Enhanced Prompt", lines=6)
phn = gr.Textbox(label="Negative Prompt", value="blurry, low quality, distorted, deformed, bad anatomy, extra limbs, watermark, signature, amateur, worst quality, low resolution", lines=2)
# Handlers
ig = ImageGen()
gb.click(fn=lambda *a: (ig.generate(*a)[0], ig.generate(*a)[1]),
inputs=[gp,gn,gw,gh,gs,gg,gd,gn_img], outputs=[go,gw_txt])
en = Enhancer()
ebu.click(fn=en.upscale, inputs=[ei,es], outputs=eo)
ebs.click(fn=en.skin, inputs=[ei,esk], outputs=eo)
pe = PoseEd()
pbx.click(fn=pe.extract, inputs=pi, outputs=po)
pbg.click(fn=lambda i,p,n: pe.views(i,p,n), inputs=[pi,pp,pn], outputs=[po,pw])
vg = VarGen()
vb.click(fn=lambda i,p,n,s: vg.gen(i,p,n,s), inputs=[vi,vp,vn,vs], outputs=[vo,vw])
vdg = VideoGen()
vdb.click(fn=vdg.generate, inputs=[vdp,vdf,vdfps,vds], outputs=[vdo,vdw])
ag = AudioGen()
amb.click(fn=ag.generate_music, inputs=[amp,amd,ams,amsz], outputs=[amo,amw])
asb.click(fn=ag.generate_sfx, inputs=[asp,asd,ass], outputs=[aso,asw])
pg = PlotGen()
plb.click(fn=pg.generate, inputs=[plg,plt,plto,pln,pls,plc], outputs=plo)
fe = FilmEd()
fpc.click(fn=fe.create, inputs=fpn, outputs=fps)
fpa.click(fn=fe.add, inputs=[fpn,fsi,fsd,fsa], outputs=fps)
fpr.click(fn=lambda p,f,t: fe.render(p,f,t), inputs=[fpn,fpf,fpt], outputs=[fpo,fprs])
def ph_enh(basic, style, lighting, quality, camera):
if not basic: return "",""
return f"{camera}, {basic}, {style}, {lighting}, {quality}, best quality, sharp focus", "blurry, low quality, distorted, deformed, bad anatomy, extra limbs, watermark, signature, amateur, worst quality, low resolution"
phbtn.click(fn=ph_enh, inputs=[phb,phst,phl,phq,phc], outputs=[pho,phn])
return demo
if __name__ == "__main__":
demo = build_ui()
demo.launch(server_name="0.0.0.0", server_port=7860)