Bria.ai pipeline modification to use Bedrock

bria-ai/fibo-bedrock/.gitignore

@@ -0,0 +1,36 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+
+# Virtual Environment
+venv/
+env/
+ENV/
+.venv/
+
+# Ignore all markdown files (including fibo-bedrock-pipe/README.md)
+*.md
+!/README.md
+
+# Ignore all Python files except specific ones
+*.py
+!fibo-bedrock-pipe/*.py
+!use_fibo.py
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+
+# OS
+.DS_Store
+Thumbs.db
+
+.json
+.png
+

bria-ai/fibo-bedrock/README.md

@@ -0,0 +1,34 @@
+---
+license: cc-by-nc-4.0
+---
+
+Bria Fibo using Amazon Bedrock 
+
+This is a modification of Bria's pipeline block that converts a prompt to a JSON object using Bedrock. See use_fibo.py for usage details.
+
+## Usage
+
+```python
+# Create a prompt to generate an initial image
+# You can also enable verbose logging per-call by adding verbose=True
+output = vlm_pipe(
+    prompt="A hyper-detailed, ultra-fluffy owl sitting in the trees at night, looking directly at the camera with wide, adorable, expressive eyes. Its feathers are soft and voluminous, catching the cool moonlight with subtle silver highlights. The owl's gaze is curious and full of charm, giving it a whimsical, storybook-like personality.",
+    # verbose=True
+)
+json_prompt_generate = output.values["json_prompt"]
+
+...
+
+results_generate = pipe(
+    prompt=json_prompt_generate, num_inference_steps=50, guidance_scale=5, negative_prompt=negative_prompt
+)
+```
+
+
+## Inputs
+
+- `prompt`: A string prompt to convert to a JSON object.
+
+## Outputs
+
+- `json_prompt`: A JSON object representing the prompt.

bria-ai/fibo-bedrock/fibo-bedrock-pipe/config.json

@@ -0,0 +1,7 @@
+{
+  "_class_name": "BriaFiboGeminiPromptToJson",
+  "_diffusers_version": "0.35.0.dev0",
+  "auto_map": {
+    "ModularPipelineBlocks": "gemini_prompt_to_json.BriaFiboGeminiPromptToJson"
+  }
+}

bria-ai/fibo-bedrock/fibo-bedrock-pipe/modular_config.json

@@ -0,0 +1,37 @@
+{
+  "_class_name": "BriaFiboGeminiPromptToJson",
+  "_diffusers_version": "0.36.0.dev0",
+  "auto_map": {
+    "ModularPipelineBlocks": "gemini_prompt_to_json.BriaFiboGeminiPromptToJson"
+  },
+  "requirements": [
+    [
+      "google.generativeai",
+      "0.8.4"
+    ],
+    [
+        "google-genai",
+        "1.46.0"
+    ],
+    [
+      "transformers",
+      "4.57.1"
+    ],
+    [
+        "pydantic",
+        "2.12.3"
+    ],
+    [
+        "boltons",
+        "25.0.0"
+    ],
+    [
+        "ujson",
+        "5.11.0"
+    ],
+    [
+        "Pillow",
+        "10.1.0"
+    ]
+  ]
+}

bria-ai/fibo-bedrock/set_models.sh

@@ -0,0 +1,15 @@
+#!/bin/bash
+# Set Bedrock model environment variables
+
+# BASE_MODEL: For text-only operations (Generate, RefineA)
+# Use efficient Nova Lite for fast, cost-effective text generation
+export BASE_MODEL="us.amazon.nova-lite-v1:0"
+
+# VLM_MODEL: For vision operations (Caption, RefineB, InspireA)  
+export VLM_MODEL="us.amazon.nova-2-lite-v1:0"
+
+echo "Environment variables set:"
+echo "  BASE_MODEL=$BASE_MODEL"
+echo "  VLM_MODEL=$VLM_MODEL"
+echo ""
+echo "Usage: source set_models.sh"

bria-ai/fibo-bedrock/use_fibo.py

@@ -0,0 +1,112 @@
+import json
+import os
+import sys
+import importlib
+
+import torch
+from diffusers import BriaFiboPipeline
+from diffusers.modular_pipelines import ModularPipeline
+
+# Add the local fibo-bedrock-pipe to the Python path
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), 'fibo-bedrock-pipe'))
+
+# Force reload of the module to avoid caching issues
+if 'bedrock_prompt_to_json' in sys.modules:
+    importlib.reload(sys.modules['bedrock_prompt_to_json'])
+
+# Import the local Bedrock-based pipeline
+from bedrock_prompt_to_json import BriaFiboBedrockPromptToJson
+
+# -------------------------------
+# Load the VLM pipeline (using local Bedrock version)
+# -------------------------------
+torch.set_grad_enabled(False)
+
+# Create the pipeline using the local Bedrock implementation
+# Model IDs are read from environment variables:
+# - BASE_MODEL: For text-only operations (default: us.amazon.nova-lite-v1:0)
+# - VLM_MODEL: For vision operations (default: us.anthropic.claude-3-5-sonnet-20241022-v2:0)
+# NOTE: presumed AWS credentials set in shell to access Bedrock
+# Set verbose=True to enable detailed logging of all operations
+blocks = BriaFiboBedrockPromptToJson(region_name="us-east-1", verbose=True)
+
+# Initialize the ModularPipeline with the local blocks
+vlm_pipe = ModularPipeline(blocks=blocks)
+
+# NOTE: Bedrock VLM uses API calls (no local models, no GPU usage)
+# All inference happens on AWS servers via boto3 client
+
+
+# Load the FIBO pipeline with aggressive memory optimizations
+# If you get a 401 error, you need to:
+# 1. Get a HuggingFace token from https://huggingface.co/settings/tokens
+# 2. Request access to the model at https://huggingface.co/briaai/FIBO
+# 3. Either login with: huggingface-cli login
+#    Or uncomment the token parameter below and add your token
+pipe = BriaFiboPipeline.from_pretrained(
+    "briaai/FIBO",
+    torch_dtype=torch.bfloat16,
+    use_safetensors=True,
+    low_cpu_mem_usage=True,  # Reduce CPU memory during loading
+    # token="hf_YOUR_TOKEN_HERE",  # Uncomment and add your token if needed
+)
+
+# Use sequential CPU offload for T4 GPU (15GB VRAM)
+# This keeps only the active component on GPU, rest on CPU
+# Slower but fits in memory
+pipe.enable_sequential_cpu_offload()
+
+# Additional memory optimizations (only use methods that BriaFiboPipeline supports)
+try:
+    pipe.enable_attention_slicing(1)  # Most aggressive slicing (slice_size=1)
+except AttributeError:
+    pass  # Not supported by this pipeline
+
+try:
+    pipe.enable_vae_slicing()  # Reduces VAE memory usage
+except AttributeError:
+    pass  # Not supported by this pipeline
+
+try:
+    pipe.enable_vae_tiling()  # Process VAE in tiles for large images
+except AttributeError:
+    pass  # Not supported by this pipeline
+
+# -------------------------------
+# Run Prompt to JSON
+# -------------------------------
+
+# Create a prompt to generate an initial image
+# You can also enable verbose logging per-call by adding verbose=True
+output = vlm_pipe(
+    prompt="A hyper-detailed, ultra-fluffy owl sitting in the trees at night, looking directly at the camera with wide, adorable, expressive eyes. Its feathers are soft and voluminous, catching the cool moonlight with subtle silver highlights. The owl's gaze is curious and full of charm, giving it a whimsical, storybook-like personality.",
+    # verbose=True  # Uncomment to enable verbose logging for this call only
+)
+json_prompt_generate = output.values["json_prompt"]
+
+def get_default_negative_prompt(existing_json: dict) -> str:
+    negative_prompt = ""
+    style_medium = existing_json.get("style_medium", "").lower()
+    if style_medium in ["photograph", "photography", "photo"]:
+        negative_prompt = """{'style_medium':'digital illustration','artistic_style':'non-realistic'}"""
+    return negative_prompt
+
+
+negative_prompt = get_default_negative_prompt(json.loads(json_prompt_generate))
+
+print("\nSaving JSON prompt to file...")
+with open("image_generate_json_prompt.json", "w") as f:
+    f.write(json_prompt_generate)
+print(f"JSON prompt saved to: image_generate_json_prompt.json")
+
+# -------------------------------
+# Run Image Generation
+# -------------------------------
+# Generate the image from the structured json prompt
+print("\nGenerating image from JSON prompt...")
+results_generate = pipe(
+    prompt=json_prompt_generate, num_inference_steps=10, guidance_scale=5, negative_prompt=negative_prompt
+)
+results_generate.images[0].save("image_generate.png")
+print(f"Image saved to: image_generate.png")
+