60 lines
1.7 KiB
Python
60 lines
1.7 KiB
Python
from fastapi import FastAPI, HTTPException
|
|
from fastapi.middleware.cors import CORSMiddleware
|
|
from pydantic import BaseModel
|
|
import torch
|
|
from diffusers import FluxPipeline
|
|
import base64
|
|
from io import BytesIO
|
|
|
|
app = FastAPI()
|
|
|
|
app.add_middleware(
|
|
CORSMiddleware,
|
|
allow_origins=["*"],
|
|
allow_credentials=True,
|
|
allow_methods=["*"],
|
|
allow_headers=["*"],
|
|
)
|
|
|
|
# Load FLUX.1-schnell (Best for 16GB VRAM + Fast Generation)
|
|
print("Initializing Flux-Schnell on RTX 5060 Ti (sm_120)...")
|
|
try:
|
|
pipe = FluxPipeline.from_pretrained(
|
|
"black-forest-labs/FLUX.1-schnell",
|
|
torch_dtype=torch.bfloat16
|
|
)
|
|
# Optimized for 16GB: Moves parts of the model to CPU when not in use
|
|
# pipe.enable_model_cpu_offload()
|
|
pipe.enable_sequential_cpu_offload()
|
|
pipe.vae.enable_tiling()
|
|
except Exception as e:
|
|
print(f"Error loading model: {e}")
|
|
|
|
class ImageRequest(BaseModel):
|
|
prompt: str
|
|
model: str = "flux-schnell" # Matches the error fix from earlier
|
|
|
|
@app.post("/v1/images/generations")
|
|
async def generate(request: ImageRequest):
|
|
try:
|
|
# Flux-Schnell generates high quality in just 4 steps
|
|
result = pipe(
|
|
request.prompt,
|
|
guidance_scale=0.0,
|
|
num_inference_steps=4,
|
|
max_sequence_length=256,
|
|
).images[0]
|
|
|
|
# Convert to Base64 for the API response
|
|
buffered = BytesIO()
|
|
result.save(buffered, format="PNG")
|
|
img_str = base64.b64encode(buffered.getvalue()).decode()
|
|
|
|
return {"data": [{"b64_json": img_str}]}
|
|
except Exception as e:
|
|
raise HTTPException(status_code=500, detail=str(e))
|
|
|
|
if __name__ == "__main__":
|
|
import uvicorn
|
|
uvicorn.run(app, host="0.0.0.0", port=5005)
|