Flux Open Schnell
A high-speed, distilled text-to-image model based on the state-of-the-art Flux Dev architecture. Comparable in quality to Flux Schnell but optimized for real-time use.
Model Overview
- Base model: Flux Dev (state-of-the-art text-to-image generator)
- Distillation: Reverse-engineered reproduction of the Flux Schnell pipeline
- Inference speed: ~3–4 seconds per image (4 sampling steps)
- Use case: Real-time and scalable applications requiring high-quality image generation
Features
- Fast: Only 4 sampling steps for near-instant generation
- Faithful: Maintains visual fidelity of the original Flux Dev
- Open & Reproducible: Full distillation pipeline published in our blog
Training & Distillation Details
Model preparation
- Loaded Flux Dev checkpoints
- Configured student–teacher setup
Distillation process
- Optimized training loop to match Flux Schnell performance
- 4 sampling-step schedule
Evaluation
- Benchmarking against Flux Schnell on standard text-to-image metrics
- Visual quality validated on diverse prompts
For the full tutorial, see our webinar:
https://www.linkedin.com/events/r-dtalk-217330157774300954624/about/
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support
Model tree for SoftServe/flux-open-schnell
Base model
black-forest-labs/FLUX.1-dev