393 lines
16 KiB
Python
393 lines
16 KiB
Python
"""
|
|
S4 Custom Image with Grok Node
|
|
Author: S4MUEL
|
|
GitHub: https://github.com/S4MUEL-404/ComfyUI-S4API
|
|
|
|
Grok image generation node with custom API key input.
|
|
"""
|
|
|
|
import json
|
|
import base64
|
|
import requests
|
|
from io import BytesIO
|
|
from typing import Optional
|
|
import torch
|
|
from PIL import Image
|
|
import numpy as np
|
|
import os
|
|
import hashlib
|
|
|
|
|
|
class S4ImageWithGrok:
|
|
"""
|
|
Generates images via Grok API endpoint.
|
|
"""
|
|
|
|
# Fixed API configuration
|
|
API_URL = "https://api.x.ai/v1/images/generations"
|
|
API_MODEL = "grok-2-image-1212"
|
|
|
|
def __init__(self):
|
|
# Ensure cache directory exists
|
|
self.cache_dir = os.path.join(os.path.dirname(os.path.dirname(__file__)), "cache")
|
|
os.makedirs(self.cache_dir, exist_ok=True)
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"prompt": (
|
|
"STRING",
|
|
{
|
|
"multiline": True,
|
|
"default": "",
|
|
"tooltip": "Text prompt for image generation with Grok",
|
|
"placeholder": "Enter your image generation prompt here..."
|
|
},
|
|
),
|
|
"api_key": (
|
|
"STRING",
|
|
{
|
|
"multiline": False,
|
|
"default": "",
|
|
"tooltip": "Your Grok API Key",
|
|
"placeholder": "xai-..."
|
|
}
|
|
),
|
|
"seed": (
|
|
"INT",
|
|
{
|
|
"default": 0,
|
|
"min": 0,
|
|
"max": 2147483647,
|
|
"step": 1,
|
|
"display": "number",
|
|
"control_after_generate": True,
|
|
"tooltip": "Random seed for reproducible results (32-bit limit for Grok API)",
|
|
},
|
|
),
|
|
},
|
|
"optional": {
|
|
"images": (
|
|
"IMAGE",
|
|
{
|
|
"default": None,
|
|
"tooltip": "Optional reference images for context"
|
|
}
|
|
),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE", "STRING")
|
|
RETURN_NAMES = ("image", "text")
|
|
FUNCTION = "generate_image"
|
|
CATEGORY = "💀PromptsO"
|
|
DESCRIPTION = "Generates images using Grok API with custom API key."
|
|
|
|
def validate_inputs(self, prompt: str, api_key: str) -> None:
|
|
"""Validate input parameters"""
|
|
if not prompt or not prompt.strip():
|
|
raise ValueError("Prompt cannot be empty")
|
|
if not api_key or not api_key.strip():
|
|
raise ValueError("API key cannot be empty")
|
|
|
|
def encode_image_to_base64(self, image_tensor: torch.Tensor) -> str:
|
|
"""Convert image tensor to base64 string"""
|
|
|
|
# Convert tensor to PIL Image
|
|
if len(image_tensor.shape) == 4:
|
|
image_tensor = image_tensor[0] # Remove batch dimension if present
|
|
|
|
# Convert from [H, W, C] to [C, H, W] and normalize
|
|
if image_tensor.shape[-1] == 3: # RGB
|
|
image_np = (image_tensor.cpu().numpy() * 255).astype(np.uint8)
|
|
image_pil = Image.fromarray(image_np)
|
|
else:
|
|
raise ValueError("Unsupported image format")
|
|
|
|
# Convert to base64
|
|
buffer = BytesIO()
|
|
image_pil.save(buffer, format="PNG")
|
|
image_b64 = base64.b64encode(buffer.getvalue()).decode("utf-8")
|
|
return image_b64
|
|
|
|
def generate_vision_cache_key(self, prompt: str, images: torch.Tensor) -> str:
|
|
"""Generate cache key for vision analysis"""
|
|
cache_data = {
|
|
"prompt": prompt.strip(),
|
|
"vision_model": "grok-2-vision-1212"
|
|
}
|
|
|
|
# Include image hash
|
|
image_hash = hashlib.md5(str(images.shape).encode() + str(images.sum().item()).encode()).hexdigest()[:8]
|
|
cache_data["images_hash"] = image_hash
|
|
|
|
cache_str = json.dumps(cache_data, sort_keys=True)
|
|
return hashlib.md5(cache_str.encode()).hexdigest()
|
|
|
|
def generate_cache_key(self, final_prompt: str, seed: int, images: Optional[torch.Tensor] = None) -> str:
|
|
"""Generate cache key based on final parameters used for generation"""
|
|
cache_data = {
|
|
"prompt": final_prompt.strip(), # Use final prompt instead of original
|
|
"model": self.API_MODEL,
|
|
"seed": seed
|
|
}
|
|
|
|
# Include image hash if images are provided (for consistency)
|
|
if images is not None:
|
|
image_hash = hashlib.md5(str(images.shape).encode() + str(images.sum().item()).encode()).hexdigest()[:8]
|
|
cache_data["images_hash"] = image_hash
|
|
|
|
cache_str = json.dumps(cache_data, sort_keys=True)
|
|
return hashlib.md5(cache_str.encode()).hexdigest()
|
|
|
|
def load_vision_cache(self, vision_cache_key: str) -> Optional[str]:
|
|
"""Load vision analysis result from cache"""
|
|
cache_file = os.path.join(self.cache_dir, f"{vision_cache_key}_vision.json")
|
|
if os.path.exists(cache_file):
|
|
try:
|
|
with open(cache_file, 'r', encoding='utf-8') as f:
|
|
cache_data = json.load(f)
|
|
return cache_data.get("enhanced_prompt")
|
|
except Exception as e:
|
|
print(f"⚠️ Failed to load vision cache: {e}")
|
|
return None
|
|
|
|
def save_vision_cache(self, vision_cache_key: str, enhanced_prompt: str) -> None:
|
|
"""Save vision analysis result to cache"""
|
|
try:
|
|
cache_file = os.path.join(self.cache_dir, f"{vision_cache_key}_vision.json")
|
|
cache_data = {
|
|
"enhanced_prompt": enhanced_prompt,
|
|
"timestamp": json.dumps({"time": "cached"})
|
|
}
|
|
with open(cache_file, 'w', encoding='utf-8') as f:
|
|
json.dump(cache_data, f, ensure_ascii=False, indent=2)
|
|
except Exception as e:
|
|
print(f"⚠️ Failed to save vision cache: {e}")
|
|
|
|
def load_from_cache(self, cache_key: str) -> Optional[tuple]:
|
|
"""Load result from cache if exists"""
|
|
cache_file = os.path.join(self.cache_dir, f"{cache_key}_grok_img.json")
|
|
if os.path.exists(cache_file):
|
|
try:
|
|
with open(cache_file, 'r', encoding='utf-8') as f:
|
|
cache_data = json.load(f)
|
|
# Load image tensor from file
|
|
img_file = os.path.join(self.cache_dir, f"{cache_key}_grok_img.pt")
|
|
if os.path.exists(img_file):
|
|
image_tensor = torch.load(img_file, weights_only=True)
|
|
return (image_tensor, cache_data.get("text", ""))
|
|
except Exception as e:
|
|
print(f"⚠️ Failed to load cache: {e}")
|
|
return None
|
|
|
|
def save_to_cache(self, cache_key: str, image_tensor: torch.Tensor, text: str) -> None:
|
|
"""Save result to cache"""
|
|
try:
|
|
# Save metadata
|
|
cache_file = os.path.join(self.cache_dir, f"{cache_key}_grok_img.json")
|
|
cache_data = {
|
|
"text": text,
|
|
"timestamp": json.dumps({"time": "cached"})
|
|
}
|
|
with open(cache_file, 'w', encoding='utf-8') as f:
|
|
json.dump(cache_data, f, ensure_ascii=False, indent=2)
|
|
|
|
# Save image tensor
|
|
img_file = os.path.join(self.cache_dir, f"{cache_key}_grok_img.pt")
|
|
torch.save(image_tensor, img_file)
|
|
except Exception as e:
|
|
print(f"⚠️ Failed to save cache: {e}")
|
|
|
|
def generate_image(
|
|
self,
|
|
prompt,
|
|
api_key,
|
|
seed=0,
|
|
images=None,
|
|
):
|
|
self.validate_inputs(prompt, api_key)
|
|
|
|
# If reference images are provided, analyze them with Grok Vision (with caching)
|
|
enhanced_prompt = prompt
|
|
if images is not None:
|
|
# Check vision cache first
|
|
vision_cache_key = self.generate_vision_cache_key(prompt, images)
|
|
cached_enhanced_prompt = self.load_vision_cache(vision_cache_key)
|
|
|
|
if cached_enhanced_prompt:
|
|
print(f"💾 Using cached vision analysis")
|
|
enhanced_prompt = cached_enhanced_prompt
|
|
else:
|
|
print(f"🖼️ Analyzing reference image(s) with Grok Vision...")
|
|
|
|
# Get image base64
|
|
if len(images.shape) == 4: # Batch of images
|
|
image_b64 = self.encode_image_to_base64(images[0])
|
|
else: # Single image
|
|
image_b64 = self.encode_image_to_base64(images)
|
|
|
|
# Use Grok Vision to analyze the image with length constraint
|
|
vision_prompt = f"Look at this image and create a short, concise image generation prompt (under 800 characters) that captures the key visual style, colors, and elements from this reference image. Incorporate this instruction: {prompt}. Focus on the most important visual aspects only."
|
|
|
|
vision_data = {
|
|
"model": "grok-2-vision-1212",
|
|
"messages": [
|
|
{
|
|
"role": "user",
|
|
"content": [
|
|
{"type": "text", "text": vision_prompt},
|
|
{
|
|
"type": "image_url",
|
|
"image_url": {"url": f"data:image/png;base64,{image_b64}"}
|
|
}
|
|
]
|
|
}
|
|
],
|
|
"max_tokens": 500,
|
|
"temperature": 0.7
|
|
}
|
|
|
|
try:
|
|
# Call Grok Vision API
|
|
vision_headers = {
|
|
"Authorization": f"Bearer {api_key}",
|
|
"Content-Type": "application/json"
|
|
}
|
|
|
|
vision_response = requests.post(
|
|
"https://api.x.ai/v1/chat/completions",
|
|
headers=vision_headers,
|
|
json=vision_data,
|
|
timeout=60
|
|
)
|
|
|
|
if vision_response.status_code == 200:
|
|
vision_result = vision_response.json()
|
|
if "choices" in vision_result and vision_result["choices"]:
|
|
enhanced_prompt = vision_result["choices"][0]["message"]["content"]
|
|
|
|
# Ensure prompt is within length limit (1024 characters for Grok)
|
|
if len(enhanced_prompt) > 1000:
|
|
# Truncate to safe length
|
|
enhanced_prompt = enhanced_prompt[:1000].rsplit(' ', 1)[0] + "..."
|
|
print(f"✂️ Enhanced prompt truncated to fit length limit")
|
|
|
|
# Save to vision cache
|
|
self.save_vision_cache(vision_cache_key, enhanced_prompt)
|
|
print(f"✨ Enhanced prompt from vision analysis ({len(enhanced_prompt)} chars)")
|
|
else:
|
|
print(f"⚠️ Vision analysis failed, using original prompt")
|
|
else:
|
|
print(f"⚠️ Vision API error: {vision_response.status_code}, using original prompt")
|
|
except Exception as e:
|
|
print(f"⚠️ Vision analysis error: {e}, using original prompt")
|
|
|
|
# Ensure final prompt is within length limit
|
|
final_prompt = enhanced_prompt
|
|
if len(final_prompt) > 1000:
|
|
# Truncate to safe length
|
|
final_prompt = final_prompt[:1000].rsplit(' ', 1)[0] + "..."
|
|
print(f"✂️ Final prompt truncated to {len(final_prompt)} characters")
|
|
|
|
# Generate cache key based on final prompt (this is the key fix)
|
|
cache_key = self.generate_cache_key(final_prompt, seed, images)
|
|
|
|
# Check cache for final result
|
|
cached_result = self.load_from_cache(cache_key)
|
|
if cached_result:
|
|
print(f"💾 Using cached image generation result for seed {seed}")
|
|
return cached_result
|
|
|
|
# Prepare request data for image generation
|
|
data = {
|
|
"prompt": final_prompt,
|
|
"n": 1,
|
|
"model": self.API_MODEL
|
|
}
|
|
|
|
# Add seed for reproducible results
|
|
if seed > 0:
|
|
data["seed"] = seed
|
|
|
|
# Make API call
|
|
headers = {
|
|
"Authorization": f"Bearer {api_key}",
|
|
"Content-Type": "application/json"
|
|
}
|
|
|
|
print(f"🤖 Generating image with Grok...")
|
|
print(f" • Model: {self.API_MODEL}")
|
|
print(f" • Original prompt: {prompt[:50]}{'...' if len(prompt) > 50 else ''}")
|
|
if images is not None:
|
|
print(f" • Reference images: {images.shape[0] if len(images.shape) == 4 else 1}")
|
|
print(f" • Enhanced prompt: {enhanced_prompt[:50]}{'...' if len(enhanced_prompt) > 50 else ''}")
|
|
print(f" • Final prompt length: {len(final_prompt)} characters")
|
|
print(f" • Seed: {seed}")
|
|
print(f" • Cache key: {cache_key[:8]}...")
|
|
|
|
try:
|
|
response = requests.post(
|
|
self.API_URL,
|
|
headers=headers,
|
|
json=data,
|
|
timeout=120
|
|
)
|
|
|
|
if response.status_code != 200:
|
|
error_data = response.json() if response.content else {}
|
|
error_msg = error_data.get("error", response.text)
|
|
raise Exception(f"API request failed with status {response.status_code}: {error_msg}")
|
|
|
|
# Parse response
|
|
result = response.json()
|
|
|
|
if "data" not in result or not result["data"]:
|
|
raise Exception("No images returned from API")
|
|
|
|
# Download and process images
|
|
image_tensors = []
|
|
response_texts = []
|
|
|
|
for i, img_data in enumerate(result["data"]):
|
|
img_url = img_data["url"]
|
|
revised_prompt = img_data.get("revised_prompt", prompt)
|
|
|
|
print(f"🔄 Downloading image {i+1}/{len(result['data'])}...")
|
|
|
|
# Download image
|
|
img_response = requests.get(img_url, timeout=60)
|
|
img_response.raise_for_status()
|
|
|
|
# Convert to PIL Image
|
|
image_pil = Image.open(BytesIO(img_response.content))
|
|
|
|
# Convert to RGB if needed
|
|
if image_pil.mode != "RGB":
|
|
image_pil = image_pil.convert("RGB")
|
|
|
|
# Convert to tensor
|
|
image_np = np.array(image_pil, dtype=np.float32) / 255.0
|
|
image_tensor = torch.from_numpy(image_np).unsqueeze(0)
|
|
image_tensors.append(image_tensor)
|
|
|
|
# Store response info
|
|
response_texts.append(f"Image {i+1}: {revised_prompt}")
|
|
|
|
# Concatenate all images
|
|
final_image_tensor = torch.cat(image_tensors, dim=0)
|
|
final_text = "\\n".join(response_texts)
|
|
|
|
print(f"✅ Images generated successfully!")
|
|
print(f" • Generated {len(image_tensors)} image(s)")
|
|
print(f" • Image tensor shape: {final_image_tensor.shape}")
|
|
|
|
# Save to cache
|
|
self.save_to_cache(cache_key, final_image_tensor, final_text)
|
|
|
|
return (final_image_tensor, final_text)
|
|
|
|
except Exception as e:
|
|
print(f"❌ Error generating image: {str(e)}")
|
|
raise e |