StoryMaker

StoryMaker is an open-source AI image generation Framework designed for creators who need strong visual continuity across a sequence of generated images — especially when charACTers must retAIn recognizable facial features, clothing, hairstyles, and body proportions from one frame to the next. Unlike Standard diffusion models that treat each Prompt in isolation, StoryMaker introduces architectural innovations to preserve IDEntity while allowing flexible background, pose, and style variation — making it uniquely suited for storyboarding, visual novels, comic prototyping, and branded content development.

What it does

StoryMaker transfORMs natural language prompts into coherent image sequences where characters remain visually stable across diverse scenes. It achieves this by integrating identity-aware conditioning: facial embeddings are extracted from reference images and fused with textual prompts via a Positional-aware Perceiver ReSampler (PPR). This enables the model to anchor character APPearance without requiring manual masking or iterative editing.

Key capabilities

• Character Consistency: Maintains facial identity, outfit, and physical traits across dozens of generated images — even when prompts change backgrounds, lighting, or camera angles.
• Multicharacter Support: Handles two or more distinct characters simultaneously, preserving each one’s unique appearance independently within shared scenes.
• Cross-Attention Control: Uses segmentation masks and MSE loss to isolate attention regions, ensuring background edits don’t distort character features.
• Pose Decoupling: LeveRAGes ControlNet-based training to separate pose guidance from identity encoding — enabling new poses without retraining or losing fidelity.
• High-Fidelity OuTPUt: Employs LoRA adapters for efficient fine-tuning, boosting resolution, texture detail, and semantic alignment without full model retraining.

Who it is for

Independent illustrators, indie game developers, visual novel writers, marketing designers, and AI researchers exploring controllable generation. It’s especially valuable for teams building nARRative-driven assets where brand or character reCognition matters — such as animated social media campaigns or interactive storytelling prototypes.

How to get started

The project is hosted publicly on GitHub. Users need technical familiarity with Python, PyTorch, and Stable Diffusion pipelines. Setup involves cloning the repository, installing dependencies, preparing reference images for character initialization, and running inference scripts with custom prompts. Pretrained weights and sample Notebooks are provided, but local GPU resources (e.g., 16GB VRAM) are recommended for optimal performance.

Pricing and access caveat

StoryMaker is fully Open Source and free to use under its GitHub license. There is no official hosted web interface or commercial SaaS veRSIon — all usage occurs locally or on self-managed infrastructure. While the core model is accessible, users should expect a moderate learning curve; documentation is community-maintained and may vary in completeness. No official support, cloud API, or mobile app exists at this time.