| Property | Value |
|---|---|
| Parameter Count | 8 billion |
| License | Apache 2.0 |
| Author | ostris |
| Model URL | https://huggingface.co/ostris/Flex.1-alpha |
| Architecture | Rectified Flow Transformer |
What is Flex.1-alpha?
Flex.1-alpha is an advanced text-to-image generation model that evolved from FLUX.1-schnell. It represents a significant advancement in open-source image generation, featuring 8 billion parameters and a unique guidance embedder that eliminates the need for traditional CFG (Classifier-Free Guidance) during image generation. The model maintains compatibility with popular inference engines while offering enhanced flexibility for fine-tuning.
Implementation Details
The model implements a rectified flow transformer architecture with 8 double transformer blocks, fewer than its predecessor FLUX.1-dev's 19 blocks. It supports a 512 token length input and includes a byprassable guidance embedder that can be operated independently of the model weights.
- Pre-trained 8B parameter base model
- Independent guidance embedder system
- Compatible with major inference engines (Diffusers, ComfyUI)
- Supports fine-tuning without guidance embedder
Core Capabilities
- Text-to-image generation with high fidelity
- True CFG capability with optional guidance embedder
- Flexible fine-tuning support
- Seamless integration with existing workflows
- 512 token context handling
Frequently Asked Questions
Q: What makes this model unique?
Flex.1-alpha's distinctive feature is its independent guidance embedder system, which allows for CFG-free image generation while maintaining the option to bypass it during fine-tuning. This makes it more versatile than similar models in the space.
Q: What are the recommended use cases?
The model is ideal for text-to-image generation tasks, particularly when fine-tuning is desired. It's especially suitable for projects requiring flexible deployment options and those that need to work within existing inference pipelines like ComfyUI or Diffusers.