HiDream Releases HiDream-O1-Image Open-Weights Image Generation Family

The evaluation methodology relies on the Artificial Analysis Elo system, which uses human preference and automated scoring to rank models based on aesthetic quality, prompt adherence, and compositional accuracy. This provides a more realistic measure of model quality than static datasets alone.

• HiDream-O1-Image 1.5 Elo: 1018 (Ranked No. 2 globally)
• Base O1-Image Elo: 979
• Dev Variant Elo: 811
• Evaluation Metric: Artificial Analysis Text-to-Image Arena

API Pricing

HiDream offers a transparent pricing structure tailored to the scale of the user's operation. For developers moving from prototyping to production, the pricing scales significantly based on the model tier selected. The 1.5 version, offering the highest fidelity, comes at a premium, while the Dev version provides an extremely low-cost entry point for testing and experimentation.

The pricing is structured per 1,000 images, making it easy for engineers to calculate the cost of large-scale dataset generation or high-volume application usage.

• HiDream-O1-Image 1.5: $80 per 1k images
• HiDream-O1-Image Base: $40 per 1k images
• HiDream-O1-Image Dev: $5 per 1k images

Use Cases

The versatility of the O1-Image family makes it suitable for a wide range of professional applications. For creative studios, the 1.5 model provides the high-fidelity assets required for concept art and marketing materials. For AI researchers, the open-weights Dev variant is an ideal foundation for training specialized LoRAs or fine-tuning on niche aesthetic styles.

Furthermore, because the model is part of a larger omnimodal ecosystem, it is uniquely suited for 'world model' applications. This includes generating consistent environments for video generation or creating high-quality training data for embodied AI agents that require realistic visual stimuli.

• High-fidelity concept art and marketing assets
• Synthetic dataset generation for ML training
• Prototyping for multimodal/video models
• Fine-tuning and research via open-weights access

Getting Started

Developers can begin integrating HiDream-O1-Image into their workflows immediately. The open-weights models are available for download on Hugging Face, allowing for local deployment using standard inference frameworks like Diffusers.

For those who prefer a managed solution, HiDream provides API access to the full suite of models, including the high-performance 1.5 version, ensuring that you can scale from a single local test to a global production environment without changing your underlying architecture.

• Access weights on Hugging Face
• Deploy locally via standard diffusion pipelines
• Scale via HiDream's managed API

API Pricing — Output: $80 per 1k images (1.5), $40 (Base), $5 (Dev) / Context: Pricing is per 1,000 images generated.

Sources