AI创作工具：梦境生成器

Install Python dependencies

uv sync

🔧 Requirements

• Python 3.11+ with uv package manager
• Ollama for prompt generation (ollama.ai)
• Hugging Face Token for gated/private model downloads only
• GPU recommended: NVIDIA (8GB+ VRAM) or Apple Silicon

🚀 Install

Option 1: Install the CLI from PyPI

Use this path if you want to run DreamGen as an installed command and do not need to edit the source code.

uv venv --python 3.11
source .venv/bin/activate  # Windows PowerShell: .venv\Scripts\Activate.ps1
uv pip install dreamgen

Verify the install without downloading image models or requiring a GPU:

dreamgen generate --mock

The plain uv pip install dreamgen command resolves from PyPI. For NVIDIA systems where you specifically want CUDA 12.4 PyTorch wheels from the PyTorch index, include the extra index during install:

uv pip install dreamgen --extra-index-url https://download.pytorch.org/whl/cu124

Option 3: Run with Docker Compose

For a production-style local run with the shipped ports and wiring:

cp .env.docker.example .env.docker
docker compose --env-file .env.docker up --build

That exposes:

• UI: http://localhost:7860
• API: http://localhost:25800
• API docs: http://localhost:25800/api/docs

For Z-Image review in Docker:

• put LoRAs under ./loras/<name>/*.safetensors
• use Settings → Models to download Z-Image-Turbo
• use Settings → Models to switch the active backend to Z-Image
• use Settings → Plugins to enable lora and then select active LoRAs in the Models panel

For Ollama-backed image generation:

• install at least one Ollama image model such as x/z-image-turbo or x/flux2-klein
• use Settings → Ollama to pick the Ollama prompt model and Ollama image model separately
• use Settings → Models to switch the active backend to Ollama Image

For Qwen-Image typography generation:

• use Settings → Models to download diffusers/qwen-image-nf4
• switch the active backend to Qwen-Image
• use prompts with explicit text such as signs, menus, posters, labels, or bilingual English/Chinese layouts
• set QWEN_IMAGE_MODEL=Qwen/Qwen-Image only on machines with enough memory for the full model
• enable QWEN_IMAGE_LIGHTNING=true only if you also want the optional Lightning LoRA few-step path

For ERNIE-Image-Turbo generation:

• use Settings → Models to download baidu/ERNIE-Image-Turbo
• switch the active backend to ERNIE-Image
• keep the prompt enhancer on for richer image prompts, or turn it off for stricter prompt following
• use prompts with visible English, Chinese, or Japanese text when you want to stress its text-in-image behavior

Option 2: Run from source

Use this path for development, local web UI work, or Docker review.

```bash

Configure the app

cp .env.example .env

Edit .env for your machine:

- OLLAMA_IMAGE_MODEL is optional and only used for IMAGE_BACKEND=ollama

- HF_TOKEN is optional for small/turbo/smoke public models

☁️ Optional: Cloudflare Hosting

DreamGen includes two Cloudflare Workers for free, global image hosting:

✨ Modern Web Interface

Beautiful, VS Code-inspired dark theme with real-time generation and organized galleries. The web interface features:

• 🎨 Smart Generation Dashboard - AI-enhanced prompts with contextual plugins
• 🖼️ Weekly Gallery Organization - Browse your creations by week with thumbnail previews
• ⚙️ Plugin Management - Configure time-aware and artistic enhancement plugins
• 📊 Real-time Status - Monitor API, GPU, and generation progress

Start the API (terminal 1)

uv run uvicorn src.api.server:app --host 127.0.0.1 --port 25800

2. Gallery API Worker (`cloudflare-gallery/`)

Purpose: Full gallery API with listing and image retrieval Use Case: Web UI backend, public gallery, API integrations

Features: - Dynamic routing via Cloudflare Pages Functions ([[path]].js) - List endpoint: GET /api/images → returns sorted image keys - Serve endpoint: GET /api/images/{path} → streams image files - R2 bucket binding: GALLERY → dreamgen-gallery - Auto-detects content types (png/jpg/webp/gif) - CORS enabled, 1-year cache for images

Setup:

cd cloudflare-gallery
npx wrangler pages deploy public

Configuration (wrangler.toml):

name = "dreamgen-gallery"
pages_build_output_dir = "public"
[[r2_buckets]]
binding = "GALLERY"
bucket_name = "dreamgen-gallery"

API Endpoints: ```bash

List prompt plugins

uv run dreamgen plugins list