README.md

---
license: apache-2.0
tags:
  - T5xxl
  - Google FLAN
---

# FLAN-T5-XXL Fused Model

**Guide (External Site):** [English](https://www.ai-image-journey.com/2025/03/flan-t5xxl-te-only.html) | [Japanese](https://note.com/ai_image_journey/n/ncc6b1c475d8f)

This repository hosts a fused version of the FLAN-T5-XXL model, created by combining the split files from [Google's FLAN-T5-XXL repository](https://huggingface.co/google/flan-t5-xxl). The files have been merged for convenience, making it easier to integrate into AI applications, including image generation workflows.

<div style="display: flex; justify-content: center; align-items: center; gap: 2em;">
  <div>
    <img src="./images/flan_t5_xxl_TE-only_FP32_sample1.png" alt="FLAN-T5-XXL sample image 1" width="400px" height="400px">
  </div>
  <div>
    <img src="./images/flan_t5_xxl_TE-only_FP32_sample2.png" alt="FLAN-T5-XXL sample image 2" width="400px" height="400px">
  </div>
</div>

Base Model: [**blue_pencil-flux1_v0.0.1**](https://huggingface.co/bluepen5805/blue_pencil-flux1)

## Key Features

- **Fused for Simplicity:** Combines split model files into a single, ready-to-use format.
- **Optimized Variants:** Available in FP32, FP16, and quantized GGUF formats to balance accuracy and resource usage.
- **Enhanced Prompt Accuracy:** Outperforms the standard T5-XXL v1.1 in generating precise outputs for image generation tasks.

## Model Variants

| Model File                          | Size   | Accuracy (SSIM Similarity) | Recommended |
|-------------------------------------|:--------:|:----------------------------:|:-------------:|
| flan_t5_xxl_full_fp32.safetensors       | 44.1GB | 100%                       |             |
| flan_t5_xxl_full_fp16.safetensors       | 22.1GB | 99.9%                      |             |
| flan_t5_xxl_TE-only_FP32.safetensors| 18.7GB | 100%                       | 🔺           |
| flan_t5_xxl_TE-only_FP16.safetensors| 9.4GB  | 99.9%                      | ✅          |
| flan_t5_xxl_TE-only_Q8_0.gguf      | 5.5GB  | 99.8%                      | ✅          |
| flan_t5_xxl_TE-only_Q6_K.gguf      | 4.4GB  | 99.7%                      | 🔺           |
| flan_t5_xxl_TE-only_Q5_K_M.gguf    | 3.8GB  | 98.4%                      | 🔺           |
| flan_t5_xxl_TE-only_Q4_K_M.gguf    | 3.2GB  | 95.2%                      |             |
| flan_t5_xxl_TE-only_Q3_K_L.gguf    | 2.6GB  | 84.9%                      |             |

### Comparison Graph

<div style="text-align: center; margin-left: auto; margin-right: auto; width: 600px; max-width: 80%;">
  <img src="./images/Flan-T5xxl MAE and SSIM Similarity.png" alt="Flan-T5xxl MAE and SSIM Similarity Graph">
</div>

For a detailed comparison, refer to [this blog post](https://www.ai-image-journey.com/2024/12/image-difference-t5xxl-clip-l.html).

## Usage Instructions

Place the downloaded model files in one of the following directories:
- `installation_folder/models/text_encoder`
- `installation_folder/models/clip`
- `installation_folder/Models/CLIP`

### Stable Diffusion WebUI Forge

In Stable Diffusion WebUI Forge, select the FLAN-T5-XXL model instead of the default T5xxl_v1_1 text encoder.

<div style="text-align: center; margin-left: auto; margin-right: auto; width: 800px; max-width: 80%;">
  <img src="./images/Screenshot of Stable Diffusion WebUI Forge text encoder selection screen.png" alt="Stable Diffusion WebUI Forge Text Encoder Selection Screen">
</div>

**Note:** Stable Diffusion WebUI Forge does not support FP32 models. Use FP16 or GGUF formats instead.

### ComfyUI

**Sample Workflow**

<div style="text-align: center; margin-left: auto; margin-right: auto; width: 1200px; max-width: 90%;">
  <img src="./images/Flux1_MultiGPU 2025.3.10.png" alt="Flux1_MultiGPU sample workflow">
</div>

<p style="text-align:center;font-size:0.8em;">This PNG image contains a ComfyUI workflow.</p>

For ComfyUI, we recommend using the [ComfyUI-MultiGPU](https://github.com/neuratech-ai/ComfyUI-MultiGPU) custom node to load the model into system RAM instead of VRAM.

<div style="text-align: center; margin-left: auto; margin-right: auto; width: 800px; max-width: 80%;">
  <img src="./images/Screenshots of ComfyUI's DualCLIPLoaderMultiGPU and DualCLIPLoaderGGUFMultiGPU custom nodes.png" alt="ComfyUI DualCLIPLoaderMultiGPU and DualCLIPLoaderGGUFMultiGPU Custom Nodes">
</div>

Use the **DualCLIPLoaderMultiGPU** or **DualCLIPLoaderGGUFMultiGPU** node and set the device to **cpu** to load the model into system RAM.

**FP32 Support:** To use FP32 text encoders in ComfyUI, launch with the `--fp32-text-enc` flag.

## Comparison: FLAN-T5-XXL vs T5-XXL v1.1

<div style="display: flex; justify-content: center; align-items: center; gap: 2em;">
  <div>
    <img src="./images/flan_t5_xxl_image.png" alt="FLAN-T5-XXL Image" width="400px" height="400px">
  </div>
  <div>
    <img src="./images/t5_xxl_v1_1_image.png" alt="T5-XXL v1.1 Image" width="400px" height="400px">
  </div>
</div>

These example images were generated using **FLAN-T5-XXL** and [**T5-XXL v1.1**](https://huggingface.co/google/t5-v1_1-xxl) models in Flux.1. FLAN-T5-XXL delivers more accurate responses to prompts.

## Further Comparisons

- [FLAN-T5-XXL vs T5-XXL v1.1](https://www.ai-image-journey.com/2024/12/clip-t5xxl-text-encoder.html)
- [FLAN-T5-XXL FP32 vs FP16 and Quantization](https://www.ai-image-journey.com/2024/12/image-difference-t5xxl-clip-l.html)

---

## License

- This model is distributed under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0).
- The uploader claims no ownership or rights over the model.