PhotoMaker V2

PhotoMaker V2 is a state-of-the-art open-source AI image generation Framework developed by Tencent ARC. designed specifically for high-fIDElity, controllable human Portrait synthesis, it builds upon the original PhotoMaker with significant improvements in identity preservation, textual control, and generation speed. Unlike General-purpose Text-to-Image models, PhotoMaker V2 specializes in generating photorealistic faces and full-body human images where subject identity remAIns stable across multiple ouTPUts — a critical capability for APPlications requiring Character Consistency, such as virtual avatars, game NPCs, or Personalized marketing assets.

What it does

PhotoMaker V2 transfORMs simple text Prompts into detailed, lifelike human images while preserving precise facial identity — even when changing poses, expressions, clothing, or backgrounds. It achieves this through a novel face-specific adapter architecture trained to inject identity Information directly into diffusion model latents. The system supports both single-reference and multi-reference personalization, enabling users to generate new images of a person using just one or a few input photos. It also integrates seamlessly with popular control mechanisms like ControlNet and T2I-Adapter, allowing fine-grained spatial guidance for pose, composition, and layout.

Key capabilities

• Identity-Consistent Generation: Maintains accurate facial features and recognizable likeness across varied scenes and styles.
• Text-guided Customization: Accepts natural language instructions to modify age, expression, attire, hairstyle, lighting, and environment without losing identity fidelity.
• Fast inference: Optimized for efficiency — produces high-resolution portraits in SEConds on consumer-grade GPUs or via cloud Inference.
• Modular Integration: ships with ready-to-use scripts for IP-Adapter-FaceID and InstantID, plus compatibility with LoRA-based fine-tuning Workflows.
• Open Weights & Reproducibility: Model weights, training code, and inference pipelines are publicly available on Hugging Face and GitHub.

Who it is for

PhotoMaker V2 serves creative Professionals, indie developers, researchers, and educators who need reliable, editable, and scalable human imagery. game studios use it for rAPId NPC prototyping; digital marketers leveRAGe it for personalized ad creatives; educators generate historical or conceptual visual aids; and AI researchers study identity preservation in generative models. Its open nature makes it especially valuable for teams building custom pipelines or benchmarking face-generation robustness.

How to get started

Users can access PhotoMaker V2 instantly via its official Hugging Face Space, which provides a no-install web interface. For Local Deployment, the GitHub repository includes Docker support, Colab Notebooks, and step-by-step setup guides for Linux, Windows, and macOS. Developers can integrate it into existing workflows using its Python API or REST endpoints. A bASIc understanding of diffusion models and GPU environments is helpful but not required for the hosted veRSIon.

Pricing and access caveat

The core PhotoMaker V2 framework is fully Open Source and free to use under the Apache 2.0 license. The official Hugging Face dEMO is Free to access without registration, though usage may be subject to rate liMITs during peak traffic. Some third-party hosted versions or commercial integrations may require credits or subscriptions — these are not controlled by Tencent ARC. Local inference requires a GPU with at least 8GB VRAM; performance may vary depending on hardware configuration and optimization level.