1.0 fast inference guide
FireRed Image Edit 1.0 Fast: Use the Official Lightning Model
Start from FireRedTeam's official 8-step Lightning checkpoint, use the published inference settings, and separate verified releases from community models called Fast.
Official Lightning checkpoint · 8 inference steps · Community naming explained
The verified fast path
Key point 01
Official checkpoint
FireRedTeam publishes FireRed-Image-Edit-1.0-Lightning-8steps-v1.0.safetensors on Hugging Face.
Key point 02
Published settings
The official Diffusers example uses 8 inference steps and true_cfg_scale set to 1.0.
Key point 03
Clear naming
A community repository containing Fast in its name is not automatically an official FireRedTeam release.
Model identity
What “FireRed Image Edit 1.0 Fast” should mean
People search for a fast 1.0 build, while the official accelerated release is named Lightning. Using that exact model name gives you a verifiable checkpoint and model card.
Official name
Use FireRed Image Edit 1.0 Lightning when you want the accelerated model released by FireRedTeam.
Eight-step run
The published example reduces inference to 8 steps; keep the matching configuration when evaluating quality and speed.
Community variants
Review a third-party model's owner, files, license, and model card independently before downloading or deploying it.
Quick start
Run the official Lightning release in five steps
Begin with the model card and a clean environment so version or dependency changes do not get mistaken for model behavior.
- 1
Open the official repository
Confirm the owner is FireRedTeam and the repository name is FireRed-Image-Edit-1.0-Lightning.
- 2
Prepare the runtime
Install the dependencies shown in the current model card and use a compatible PyTorch, Diffusers, and CUDA environment.
- 3
Load the Lightning weights
Download the official 8-step safetensors file or let the documented pipeline resolve the repository files.
- 4
Match the settings
Start with num_inference_steps=8 and true_cfg_scale=1.0 before experimenting with other values.
- 5
Benchmark your own edit
Use the same input, prompt, seed, resolution, and hardware when comparing the base and accelerated paths.
Comparison
Official Lightning versus community “Fast” models
The names may suggest the same goal, but provenance and recommended settings are not interchangeable.
| Comparison | Official 1.0 Lightning | Community Fast variant |
|---|---|---|
| Publisher | FireRedTeam | Independent repository owner |
| Verified filename | Lightning-8steps-v1.0.safetensors | Depends on the repository |
| Published steps | 8 in the official example | Check its own model card |
| Support source | Official model card and GitHub | Maintainer documentation |
| Recommended trust check | Verify FireRedTeam owner | Audit owner, files, license, and code |
Official settings
Keep the 8-step parameters visible in code
This focused excerpt shows the two published acceleration settings. Use the full pipeline example from the current official model card for imports and image handling.
- Load the checkpoint from the official FireRedTeam Lightning repository.
- Keep seed, image size, and hardware constant when measuring speed.
- Recheck the upstream model card when dependencies or examples change.
result = pipe(
image=input_image,
prompt=prompt,
num_inference_steps=8,
true_cfg_scale=1.0,
).images[0]Troubleshooting
Avoid misleading speed tests and setup failures
Fast inference still depends on model provenance, compatible dependencies, memory, warm-up, and repeatable benchmark conditions.
Downloaded the wrong model
Check that the repository owner is FireRedTeam and the filename contains Lightning-8steps-v1.0 before loading it.
Pipeline arguments are rejected
Align your Diffusers and project dependencies with the versions documented by the current official example.
The first run is slow
Separate model download, pipeline loading, compilation, and warm-up from steady-state inference timing.
Quality differs from examples
Restore the official 8-step and CFG settings, then match seed, resolution, prompt detail, and source-image quality.
Primary sources
Use official pages for weights and parameters
These links establish the accelerated model's publisher, base release, and reference implementation.
- FireRed Image Edit 1.0 Lightning
Official accelerated model page with the 8-step checkpoint and usage example.
- FireRed Image Edit 1.0
Official base-model page for the original 1.0 release.
- FireRed Image Edit GitHub
Official project code, requirements, inference commands, and documentation.
FAQ
FireRed Image Edit 1.0 Fast questions
Is FireRed Image Edit 1.0 Fast an official model name?
The verified official accelerated release is named FireRed Image Edit 1.0 Lightning. Some community repositories use Fast in their names, so always check the owner.
How many steps does the official Lightning model use?
The FireRedTeam Diffusers example uses num_inference_steps=8 with true_cfg_scale=1.0.
What is the official checkpoint filename?
The official model page provides FireRed-Image-Edit-1.0-Lightning-8steps-v1.0.safetensors.
Is an 8-step run always the same speed?
No. Runtime varies with GPU, precision, offloading, image resolution, warm-up, software versions, and surrounding pipeline work.
Should I use Lightning or the base 1.0 model?
Use Lightning when lower step count and faster iteration matter. Compare against the base model on your own edits when maximum quality is more important.
Can I try FireRed without downloading the model?
Yes. The browser editor lets you test image-edit instructions before committing to a local model setup.
Related resources
Choose your next FireRed Image Edit guide
FireRed Image Edit 1.1
Review the latest model, online demo, benchmark, GitHub, and Hugging Face links.
FireRed Image Edit ComfyUI
Install the official 1.0 workflow JSON and fix common node or model-path errors.
Complete FireRed guide
Learn local installation, inference commands, prompt patterns, and troubleshooting.
Test before setup
Try a FireRed edit in your browser
Validate your source image and instruction online, then reproduce the edit with the official Lightning model locally.