Spaces:
Running
Running
File size: 6,940 Bytes
cd35cc5 cd40a43 cd35cc5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 |
"""
Modal serverless GPU client for Flux/SDXL image generation.
Handles: Open-source image generation models on demand.
"""
import os
from typing import Optional
import httpx
import base64
# Note: For full Modal integration, you'd deploy a Modal app.
# This client calls a deployed Modal endpoint or falls back to HuggingFace.
class ModalFluxClient:
"""Modal-powered Flux/SDXL image generation for Pip."""
# New HuggingFace router API (the old api-inference.huggingface.co is deprecated)
HF_ROUTER_URL = "https://router.huggingface.co"
# Router endpoints for different models via fal.ai
ROUTER_ENDPOINTS = {
"flux": "/fal-ai/fal-ai/flux/schnell",
"flux_dev": "/fal-ai/fal-ai/flux/dev",
}
# Legacy models (for Modal deployment)
MODELS = {
"flux": "black-forest-labs/FLUX.1-schnell",
"sdxl_lightning": "ByteDance/SDXL-Lightning",
"sdxl": "stabilityai/stable-diffusion-xl-base-1.0",
}
def __init__(self):
self.hf_token = os.getenv("HF_TOKEN")
self.modal_endpoint = os.getenv("MODAL_FLUX_ENDPOINT") # If deployed
self.available = bool(self.hf_token) or bool(self.modal_endpoint)
if not self.available:
print("⚠️ HuggingFace/Modal: No tokens found - image generation limited")
def is_available(self) -> bool:
"""Check if the client is available."""
return self.available
async def generate_image(
self,
prompt: str,
model: str = "flux"
) -> Optional[str]:
"""
Generate image using Flux or SDXL via Modal/HuggingFace Router.
Returns base64 encoded image.
"""
# Try Modal endpoint first if available
if self.modal_endpoint:
result = await self._generate_via_modal(prompt, model)
if result:
return result
# Try new HuggingFace router API (primary method)
result = await self._generate_via_hf_router(prompt, model)
if result:
return result
# Final fallback - return None
print(f"All image generation methods failed for model: {model}")
return None
async def _generate_via_modal(self, prompt: str, model: str) -> Optional[str]:
"""
Call deployed Modal function for image generation.
"""
try:
async with httpx.AsyncClient(timeout=60.0) as client:
response = await client.post(
self.modal_endpoint,
json={"prompt": prompt, "model": model}
)
if response.status_code == 200:
data = response.json()
return data.get("image_base64")
except Exception as e:
print(f"Modal generation error: {e}")
return None
async def _generate_via_hf_router(
self,
prompt: str,
model: str = "flux"
) -> Optional[str]:
"""
Generate image via new HuggingFace Router API (fal.ai backend).
This is the current working method as of 2025.
"""
try:
# Get router endpoint for model
endpoint = self.ROUTER_ENDPOINTS.get(model, self.ROUTER_ENDPOINTS["flux"])
url = f"{self.HF_ROUTER_URL}{endpoint}"
headers = {}
if self.hf_token:
headers["Authorization"] = f"Bearer {self.hf_token}"
# New API uses 'prompt' not 'inputs'
payload = {"prompt": prompt}
async with httpx.AsyncClient(timeout=120.0) as client:
response = await client.post(
url,
headers=headers,
json=payload
)
if response.status_code == 200:
data = response.json()
# New format returns {"images": [{"url": "...", "content_type": "..."}], ...}
if "images" in data and data["images"]:
image_info = data["images"][0]
# Image could be URL or base64
if isinstance(image_info, dict):
if "url" in image_info:
# Download image from URL and convert to base64
img_response = await client.get(image_info["url"])
if img_response.status_code == 200:
return base64.b64encode(img_response.content).decode('utf-8')
elif "b64_json" in image_info:
return image_info["b64_json"]
elif isinstance(image_info, str):
# Direct base64 string
return image_info
print(f"HF Router unexpected response format: {list(data.keys())}")
else:
print(f"HF Router API error: {response.status_code} - {response.text[:200]}")
except Exception as e:
print(f"HF Router generation error: {e}")
return None
async def generate_fast(self, prompt: str) -> Optional[str]:
"""
Use fastest available model (SDXL-Lightning).
"""
return await self.generate_image(prompt, model="sdxl_lightning")
async def generate_artistic(self, prompt: str) -> Optional[str]:
"""
Use Flux for more artistic, dreamlike results.
"""
return await self.generate_image(prompt, model="flux")
# Modal app definition for deployment (optional)
# Run with: modal deploy services/modal_flux.py
MODAL_APP_CODE = '''
import modal
app = modal.App("pip-flux-generator")
# Define the image with required dependencies
flux_image = modal.Image.debian_slim().pip_install(
"diffusers",
"transformers",
"accelerate",
"torch",
"safetensors"
)
@app.function(
image=flux_image,
gpu="A10G",
timeout=300,
)
def generate_flux_image(prompt: str) -> bytes:
"""Generate image using Flux on Modal GPU."""
import torch
from diffusers import FluxPipeline
pipe = FluxPipeline.from_pretrained(
"black-forest-labs/FLUX.1-schnell",
torch_dtype=torch.bfloat16
)
pipe.to("cuda")
image = pipe(
prompt,
guidance_scale=0.0,
num_inference_steps=4,
max_sequence_length=256,
).images[0]
# Convert to bytes
import io
buf = io.BytesIO()
image.save(buf, format="PNG")
return buf.getvalue()
@app.local_entrypoint()
def main(prompt: str = "a serene lake at sunset"):
image_bytes = generate_flux_image.remote(prompt)
with open("output.png", "wb") as f:
f.write(image_bytes)
print("Image saved to output.png")
'''
|