Commit ·
676dc08
1
Parent(s): 8fe167c
fix: replace AutoPipelineForText2Video with DiffusionPipeline and add custom AI Mind Settings to prompt composer
Browse files
app.py
CHANGED
|
@@ -63,18 +63,24 @@ def get_device_info():
|
|
| 63 |
|
| 64 |
DEVICE = get_device_info()
|
| 65 |
|
| 66 |
-
def enhance_prompt_with_ai(prompt_text, style_preset):
|
| 67 |
if not prompt_text or not prompt_text.strip():
|
| 68 |
return "Please enter a prompt first."
|
| 69 |
|
| 70 |
-
url = "https://invictatill-invictatill-ai.hf.space/api/v1/chat"
|
| 71 |
-
|
| 72 |
style_modifiers = STYLE_PRESETS.get(style_preset, "")
|
| 73 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
system_prompt = (
|
| 75 |
"You are an expert cinematic prompt engineer for video generation models (like Wan 2.1, LTX-Video, CogVideo). "
|
| 76 |
"Your task is to rewrite the user's simple prompt into a highly descriptive, visually stunning, "
|
| 77 |
"and detailed prompt optimized for text-to-video models. "
|
|
|
|
| 78 |
"Include specific details about lighting, camera angle, motion, and atmosphere. "
|
| 79 |
"Keep it under 75 words. "
|
| 80 |
"Respond ONLY with the final enhanced prompt. Do NOT include any intro or conversational filler."
|
|
@@ -87,9 +93,15 @@ def enhance_prompt_with_ai(prompt_text, style_preset):
|
|
| 87 |
payload = {
|
| 88 |
"message": f"[SYSTEM CONTEXT]\n{system_prompt}\n\nUser: {full_prompt}\n\nAssistant:"
|
| 89 |
}
|
| 90 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 91 |
try:
|
| 92 |
-
response = requests.post(
|
| 93 |
if response.status_code == 200:
|
| 94 |
data = response.json()
|
| 95 |
enhanced = data.get("reply") or data.get("choices", [{}])[0].get("message", {}).get("content", "")
|
|
@@ -116,26 +128,27 @@ def load_model(model_name: str, progress=gr.Progress()):
|
|
| 116 |
progress(0.2, desc=f"Downloading/Loading {model_name}...")
|
| 117 |
|
| 118 |
try:
|
|
|
|
| 119 |
if "wan" in model_id.lower():
|
| 120 |
try:
|
| 121 |
from diffusers import WanPipeline
|
| 122 |
pipe = WanPipeline.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
| 123 |
except ImportError:
|
| 124 |
-
|
| 125 |
-
pipe = AutoPipelineForText2Video.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
| 126 |
elif "ltx" in model_id.lower():
|
| 127 |
try:
|
| 128 |
from diffusers import LTXVideoPipeline
|
| 129 |
pipe = LTXVideoPipeline.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
| 130 |
except ImportError:
|
| 131 |
-
|
| 132 |
-
pipe = AutoPipelineForText2Video.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
| 133 |
elif "cogvideo" in model_id.lower():
|
| 134 |
-
|
| 135 |
-
|
|
|
|
|
|
|
|
|
|
| 136 |
else:
|
| 137 |
-
|
| 138 |
-
pipe = AutoPipelineForText2Video.from_pretrained(
|
| 139 |
model_id,
|
| 140 |
torch_dtype=DEVICE["dtype"],
|
| 141 |
variant="fp16" if DEVICE["device"] == "cuda" else None,
|
|
@@ -387,8 +400,27 @@ with gr.Blocks(css=custom_css, title="InvictaTill VideoGen Studio", theme=gr.the
|
|
| 387 |
gr.Markdown("### ✍️ Prompt Composer")
|
| 388 |
prompt = gr.Textbox(label="Describe your scene", placeholder="A cyberpunk drone shot flying through neon-lit Tokyo streets...", lines=4, elem_id="prompt")
|
| 389 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 390 |
with gr.Row():
|
| 391 |
-
enhance_btn = gr.Button("✨ Enhance Prompt with InvictaTill AI", variant="secondary")
|
| 392 |
|
| 393 |
negative_prompt = gr.Textbox(label="Negative Prompt", placeholder="blur, distortion, low quality, watermark", lines=2, value="blur, distortion, low quality, watermark, text, bad anatomy, deformed, cartoonish, static")
|
| 394 |
|
|
@@ -401,7 +433,7 @@ with gr.Blocks(css=custom_css, title="InvictaTill VideoGen Studio", theme=gr.the
|
|
| 401 |
# Click Handlers
|
| 402 |
enhance_btn.click(
|
| 403 |
fn=enhance_prompt_with_ai,
|
| 404 |
-
inputs=[prompt, style_dropdown],
|
| 405 |
outputs=[prompt]
|
| 406 |
)
|
| 407 |
|
|
|
|
| 63 |
|
| 64 |
DEVICE = get_device_info()
|
| 65 |
|
| 66 |
+
def enhance_prompt_with_ai(prompt_text, style_preset, ai_url, ai_key, session_id):
|
| 67 |
if not prompt_text or not prompt_text.strip():
|
| 68 |
return "Please enter a prompt first."
|
| 69 |
|
|
|
|
|
|
|
| 70 |
style_modifiers = STYLE_PRESETS.get(style_preset, "")
|
| 71 |
|
| 72 |
+
# Standardize URL
|
| 73 |
+
ai_url = (ai_url or "").strip().rstrip("/")
|
| 74 |
+
if not ai_url:
|
| 75 |
+
ai_url = "https://invictatill-invictatill-ai.hf.space"
|
| 76 |
+
|
| 77 |
+
chat_endpoint = f"{ai_url}/api/v1/chat"
|
| 78 |
+
|
| 79 |
system_prompt = (
|
| 80 |
"You are an expert cinematic prompt engineer for video generation models (like Wan 2.1, LTX-Video, CogVideo). "
|
| 81 |
"Your task is to rewrite the user's simple prompt into a highly descriptive, visually stunning, "
|
| 82 |
"and detailed prompt optimized for text-to-video models. "
|
| 83 |
+
"Use your learned facts, memory context, and knowledge about the user's business if relevant. "
|
| 84 |
"Include specific details about lighting, camera angle, motion, and atmosphere. "
|
| 85 |
"Keep it under 75 words. "
|
| 86 |
"Respond ONLY with the final enhanced prompt. Do NOT include any intro or conversational filler."
|
|
|
|
| 93 |
payload = {
|
| 94 |
"message": f"[SYSTEM CONTEXT]\n{system_prompt}\n\nUser: {full_prompt}\n\nAssistant:"
|
| 95 |
}
|
| 96 |
+
if session_id:
|
| 97 |
+
payload["session_id"] = session_id
|
| 98 |
+
|
| 99 |
+
headers = {"Content-Type": "application/json"}
|
| 100 |
+
if ai_key:
|
| 101 |
+
headers["Authorization"] = f"Bearer {ai_key}"
|
| 102 |
+
|
| 103 |
try:
|
| 104 |
+
response = requests.post(chat_endpoint, json=payload, headers=headers, timeout=12)
|
| 105 |
if response.status_code == 200:
|
| 106 |
data = response.json()
|
| 107 |
enhanced = data.get("reply") or data.get("choices", [{}])[0].get("message", {}).get("content", "")
|
|
|
|
| 128 |
progress(0.2, desc=f"Downloading/Loading {model_name}...")
|
| 129 |
|
| 130 |
try:
|
| 131 |
+
from diffusers import DiffusionPipeline
|
| 132 |
if "wan" in model_id.lower():
|
| 133 |
try:
|
| 134 |
from diffusers import WanPipeline
|
| 135 |
pipe = WanPipeline.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
| 136 |
except ImportError:
|
| 137 |
+
pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
|
|
|
| 138 |
elif "ltx" in model_id.lower():
|
| 139 |
try:
|
| 140 |
from diffusers import LTXVideoPipeline
|
| 141 |
pipe = LTXVideoPipeline.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
| 142 |
except ImportError:
|
| 143 |
+
pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
|
|
|
| 144 |
elif "cogvideo" in model_id.lower():
|
| 145 |
+
try:
|
| 146 |
+
from diffusers import CogVideoXPipeline
|
| 147 |
+
pipe = CogVideoXPipeline.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
| 148 |
+
except ImportError:
|
| 149 |
+
pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=DEVICE["dtype"])
|
| 150 |
else:
|
| 151 |
+
pipe = DiffusionPipeline.from_pretrained(
|
|
|
|
| 152 |
model_id,
|
| 153 |
torch_dtype=DEVICE["dtype"],
|
| 154 |
variant="fp16" if DEVICE["device"] == "cuda" else None,
|
|
|
|
| 400 |
gr.Markdown("### ✍️ Prompt Composer")
|
| 401 |
prompt = gr.Textbox(label="Describe your scene", placeholder="A cyberpunk drone shot flying through neon-lit Tokyo streets...", lines=4, elem_id="prompt")
|
| 402 |
|
| 403 |
+
with gr.Accordion("🧠 InvictaTill AI Mind Integration Settings", open=False):
|
| 404 |
+
gr.Markdown("Configure the endpoint URL and API Key for your running InvictaTill AI Space instance so the prompt enhancer can read your business insights and custom memories.")
|
| 405 |
+
ai_url_input = gr.Textbox(
|
| 406 |
+
value=os.environ.get("VITE_INVICTATILL_AI_URL", "https://invictatill-invictatill-ai.hf.space"),
|
| 407 |
+
label="AI Mind Endpoint URL",
|
| 408 |
+
placeholder="https://invictatill-invictatill-ai.hf.space"
|
| 409 |
+
)
|
| 410 |
+
ai_key_input = gr.Textbox(
|
| 411 |
+
value=os.environ.get("VITE_INVICTATILL_AI_KEY", ""),
|
| 412 |
+
label="API Key / Auth Token",
|
| 413 |
+
placeholder="invicta_sk_...",
|
| 414 |
+
type="password"
|
| 415 |
+
)
|
| 416 |
+
session_id_input = gr.Textbox(
|
| 417 |
+
value="videogen_studio_session",
|
| 418 |
+
label="Session ID (Loads Memory Context)",
|
| 419 |
+
placeholder="videogen_studio_session"
|
| 420 |
+
)
|
| 421 |
+
|
| 422 |
with gr.Row():
|
| 423 |
+
enhance_btn = gr.Button("✨ Enhance Prompt with InvictaTill AI Mind", variant="secondary")
|
| 424 |
|
| 425 |
negative_prompt = gr.Textbox(label="Negative Prompt", placeholder="blur, distortion, low quality, watermark", lines=2, value="blur, distortion, low quality, watermark, text, bad anatomy, deformed, cartoonish, static")
|
| 426 |
|
|
|
|
| 433 |
# Click Handlers
|
| 434 |
enhance_btn.click(
|
| 435 |
fn=enhance_prompt_with_ai,
|
| 436 |
+
inputs=[prompt, style_dropdown, ai_url_input, ai_key_input, session_id_input],
|
| 437 |
outputs=[prompt]
|
| 438 |
)
|
| 439 |
|