prithivMLmods commited on
Commit
ade7305
·
verified ·
1 Parent(s): d30f10c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -1
README.md CHANGED
@@ -11,6 +11,10 @@ tags:
11
  - 8bit
12
  ---
13
 
 
 
 
 
14
  ## Quick Start with Diffusers 🧨
15
 
16
  ### Install the required packages
@@ -219,4 +223,7 @@ with gr.Blocks() as demo:
219
 
220
  if __name__ == "__main__":
221
  demo.queue(max_size=30).launch(css=css, mcp_server=True, ssr_mode=False, show_error=True)
222
- ```
 
 
 
 
11
  - 8bit
12
  ---
13
 
14
+ # **FireRed-Image-Edit-1.0-8bit**
15
+
16
+ > FireRed-Image-Edit-1.0-8bit is an 8-bit quantized edition of FireRed-Image-Edit-1.0 (FireRedTeam), engineered to deliver the same instruction-driven diffusion transformer image editing capabilities with significantly reduced memory footprint and improved inference efficiency. Built upon the original 1.6B-sample training corpus refined into over 100M high-quality text-to-image and editing pairs through cleaning, stratification, auto-labeling, and dual-stage semantic filtering, this quantized release preserves the model’s multi-stage training pipeline, including large-scale pre-training, supervised fine-tuning, and reinforcement learning with techniques such as Multi-Condition Aware Bucket Sampling for variable resolutions, Stochastic Instruction Alignment, Asymmetric Gradient Optimization for stable DPO, DiffusionNFT with layout-OCR rewards for precise text editing, and differentiable Consistency Loss for strong identity preservation. The 8-bit quantization reduces VRAM requirements and accelerates deployment while maintaining high alignment, semantic consistency, and visual fidelity across diverse editing scenarios such as photo restoration, object insertion and modification, style transfer with text fidelity, multi-image virtual try-on, and layout-aware text editing. Optimized for practical workflows and ComfyUI integration, this version enables broader accessibility on consumer-grade GPUs without substantial quality degradation, making it suitable for research, production, and lightweight deployment environments.
17
+
18
  ## Quick Start with Diffusers 🧨
19
 
20
  ### Install the required packages
 
223
 
224
  if __name__ == "__main__":
225
  demo.queue(max_size=30).launch(css=css, mcp_server=True, ssr_mode=False, show_error=True)
226
+ ```
227
+
228
+ > [!IMPORTANT]
229
+ This repository follows the same release notes, terms and conditions, and license as the original model page, FireRed-Image-Edit-1.0.