henrisel/DeltaVR3DModelGeneration
1
0
Fork 0
forked from cgvr/DeltaVR
Code Pull Requests Activity

integrate local trellis api into start_pipeline.py

Browse Source
This commit is contained in:
henrisel 2025-11-07 16:22:13 +02:00
parent 447449e1b3
commit 09f764c0df
6 changed files with 183 additions and 94 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
3d-generation-pipeline/.env.example
View File

@ -1,4 +1,8 @@
PIPELINE_FOLDER=
REFINE_PROMPT=0
CLOUDFLARE_ACCOUNT_ID=
CLOUDFLARE_API_TOKEN=
PIPELINE_FOLDER=
MODEL_FOLDER=
3D_GENERATION_URL=
MODEL_FOLDER=

1
3d-generation-pipeline/.gitignore vendored
View File

@ -1,4 +1,5 @@
.venv
.env
__pycache__
images/
models/

59
3d-generation-pipeline/cloudflare_api.py Normal file
View File

@ -0,0 +1,59 @@
import base64
import requests
import os
from dotenv import load_dotenv
load_dotenv()
ACCOUNT_ID = os.environ["CLOUDFLARE_ACCOUNT_ID"]
API_TOKEN = os.environ["CLOUDFLARE_API_TOKEN"]
def text_to_image(prompt, output_path):
MODEL = "@cf/black-forest-labs/flux-1-schnell"
URL = f"https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/{MODEL}"
payload = {
"prompt": prompt,
}
headers = {
"Authorization": f"Bearer {API_TOKEN}",
"Content-Type": "application/json",
}
resp = requests.post(URL, json=payload, headers=headers, timeout=60)
resp.raise_for_status()
data = resp.json()
b64 = data["result"]["image"]
if not b64:
raise RuntimeError(f"Unexpected response structure: {data}")
img_bytes = base64.b64decode(b64)
with open(output_path, "wb") as f:
f.write(img_bytes)
def refine_text_prompt(prompt):
MODEL = "@cf/meta/llama-3.2-3b-instruct"
URL = f"https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/{MODEL}"
instructions = """
User is talking about some object. Your task is to generate a short and concise description of it. Use only user's own words, keep it as short as possible.
Example:
User: 'Umm, okay, I would like a really cool sword, with for example a bright orange crossguard. And also it should be slightly curved.'
You: 'a slightly curved sword with bright orange crossguard'
"""
response = requests.post(URL,
headers={"Authorization": f"Bearer {API_TOKEN}"},
json={
"messages": [
{"role": "system", "content": instructions},
{"role": "user", "content": prompt}
]
}
)
data = response.json()
return data["result"]["response"]

92
3d-generation-pipeline/generate_model_local.py Normal file
View File

@ -0,0 +1,92 @@
import subprocess
import os
import time
import requests
import base64
from dotenv import load_dotenv
load_dotenv()
MODEL_FOLDER = os.environ["MODEL_FOLDER"]
API_URL = os.environ["3D_GENERATION_URL"]
def image_to_3d_subprocess(image_path, output_path):
venv_python = MODEL_FOLDER + r"\.venv\Scripts\python.exe"
script_path = MODEL_FOLDER + r"\run.py"
args = [image_path, "--output-dir", output_path]
command = [venv_python, script_path] + args
try:
# Run the subprocess
result = subprocess.run(command, capture_output=True, text=True)
# Print output and errors
print("STDOUT:\n", result.stdout)
print("STDERR:\n", result.stderr)
print("Return Code:", result.returncode)
except Exception as e:
print(f"Error occurred: {e}")
def generate_no_preview(image_base64: str):
"""Generate 3D model from a single base64-encoded image without previews.
Args:
image_base64: Base64 string of the image (without 'data:image/...' prefix)
"""
try:
# Set generation parameters
params = {
'image_base64': image_base64,
'seed': 42,
'ss_guidance_strength': 7.5,
'ss_sampling_steps': 30,
'slat_guidance_strength': 7.5,
'slat_sampling_steps': 30,
'mesh_simplify_ratio': 0.95,
'texture_size': 1024,
'output_format': 'glb'
}
# Start generation
print("Starting generation...")
response = requests.post(f"{API_URL}/generate_no_preview", data=params)
response.raise_for_status()
# Poll status until complete
while True:
status = requests.get(f"{API_URL}/status").json()
print(f"Progress: {status['progress']}%")
if status['status'] == 'COMPLETE':
break
elif status['status'] == 'FAILED':
raise Exception(f"Generation failed: {status['message']}")
time.sleep(1)
# Download the model
print("Downloading model...")
response = requests.get(f"{API_URL}/download/model")
response.raise_for_status()
return response.content
except Exception as e:
print(f"Error: {str(e)}")
return None
def image_to_3d_api(image_path, output_path):
with open(image_path, 'rb') as image_file:
image_data = image_file.read()
base64_encoded = base64.b64encode(image_data).decode('utf-8')
model_binary = generate_no_preview(base64_encoded)
with open(output_path, 'wb') as f:
f.write(model_binary)

16
3d-generation-pipeline/notebooks/local_model_generation_test.ipynb
View File

File diff suppressed because one or more lines are too long

101
3d-generation-pipeline/start_pipeline.py
View File

@ -1,111 +1,46 @@
import os
import base64
import requests
import argparse
import subprocess
from pathlib import Path
from datetime import datetime
from dotenv import load_dotenv
from cloudflare_api import text_to_image, refine_text_prompt
from generate_model_local import image_to_3d_api, image_to_3d_subprocess
load_dotenv()
ACCOUNT_ID = os.environ["CLOUDFLARE_ACCOUNT_ID"]
API_TOKEN = os.environ["CLOUDFLARE_API_TOKEN"]
PIPELINE_FOLDER = os.environ["PIPELINE_FOLDER"]
MODEL_FOLDER = os.environ["MODEL_FOLDER"]
def get_timestamp():
return datetime.now().strftime("%Y-%m-%d-%H-%M-%S")
def text_to_image(prompt, output_path):
MODEL = "@cf/black-forest-labs/flux-1-schnell"
URL = f"https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/{MODEL}"
payload = {
"prompt": prompt,
}
headers = {
"Authorization": f"Bearer {API_TOKEN}",
"Content-Type": "application/json",
}
resp = requests.post(URL, json=payload, headers=headers, timeout=60)
resp.raise_for_status()
data = resp.json()
b64 = data["result"]["image"]
if not b64:
raise RuntimeError(f"Unexpected response structure: {data}")
img_bytes = base64.b64decode(b64)
with open(output_path, "wb") as f:
f.write(img_bytes)
def refine_text_prompt(prompt):
MODEL = "@cf/meta/llama-3.2-3b-instruct"
URL = f"https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/{MODEL}"
instructions = """
User is talking about some object. Your task is to generate a short and concise description of it. Use only user's own words, keep it as short as possible.
Example:
User: 'Umm, okay, I would like a really cool sword, with for example a bright orange crossguard. And also it should be slightly curved.'
You: 'a slightly curved sword with bright orange crossguard'
"""
response = requests.post(URL,
headers={"Authorization": f"Bearer {API_TOKEN}"},
json={
"messages": [
{"role": "system", "content": instructions},
{"role": "user", "content": prompt}
]
}
)
data = response.json()
return data["result"]["response"]
def image_to_3d(image_path, output_path):
venv_python = MODEL_FOLDER + r"\.venv\Scripts\python.exe"
script_path = MODEL_FOLDER + r"\run.py"
args = [image_path, "--output-dir", output_path]
command = [venv_python, script_path] + args
try:
# Run the subprocess
result = subprocess.run(command, capture_output=True, text=True)
# Print output and errors
print("STDOUT:\n", result.stdout)
print("STDERR:\n", result.stderr)
print("Return Code:", result.returncode)
except Exception as e:
print(f"Error occurred: {e}")
def main():
parser = argparse.ArgumentParser(description="Text to 3D model pipeline")
parser.add_argument("--prompt", type=str, required=True, help="User text prompt")
args = parser.parse_args()
user_prompt = args.prompt
print(f"User prompt: {user_prompt}")
refined_prompt = refine_text_prompt(user_prompt)
print(f"Refined prompt: {refined_prompt}")
input_prompt = args.prompt
print(f"Input prompt: {input_prompt}")
refine_prompt = os.environ["REFINE_PROMPT"] == "1"
if refine_prompt:
image_generation_prompt = refine_text_prompt(input_prompt)
print(f"Refined prompt: {image_generation_prompt}")
else:
image_generation_prompt = input_prompt
timestamp = get_timestamp()
pipeline_folder = Path(PIPELINE_FOLDER)
image_path = pipeline_folder / "images" / f"{timestamp}.jpg"
text_to_image(refined_prompt, image_path)
text_to_image(image_generation_prompt, image_path)
print(f"Generated image file: {image_path}")
model_path = pipeline_folder / "models" / timestamp
image_to_3d(image_path, model_path)
model_file_path = model_path / "0" / "mesh.glb"
print(f"Generated 3D model file: {model_file_path}")
image_to_3d_api(image_path, model_path)
#model_file_path = model_path / "0" / "mesh.glb"
print(f"Generated 3D model file: {model_path}")
if __name__ == "__main__":