FLAN-T5-XXL Fused Model

Update History

April 20, 2025

Updated Stable Diffusion WebUI Forge FP32 launch argument.

April 15, 2025

Updated content to reflect ComfyUI updates.

March 20, 2025

Updated FLAN-T5-XXL model list and table.

Guide (External Site): English | Japanese

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. The files have been merged for convenience, making it easier to integrate into AI applications, including image generation workflows.

Base Model: blue_pencil-flux1_v0.0.1

Key Features

Fused for Simplicity: Combines split model files into a single, ready-to-use format.
Optimized Variants: Available in FP32, FP16, FP8, 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	Size	SSIM Similarity	Recommended
FP32	19 GB	100.0%	🔺
FP16	9.6 GB	98.0%	✅
FP8	4.8 GB	95.3%	🔺
Q8_0	6 GB	97.6%	✅
Q6_K	4.9 GB	97.3%	🔺
Q5_K_M	4.3 GB	94.8%
Q4_K_M	3.7 GB	96.4%

Comparison Graph

FLAN-T5-XXL MAE and SSIM Similarity Graph

For a detailed comparison, refer to this blog post.

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

ComfyUI

When using Flux.1 in ComfyUI, load the text encoder with the DualCLIPLoader node.

As of April 13, 2025, the default DualCLIPLoader node includes a device selection option, allowing you to choose where to load the model:

cuda → VRAM
cpu → System RAM

Since Flux.1’s text encoder is large, setting the device to cpu and storing the model in system RAM often improves performance. Unless your system RAM is 16GB or less, keeping the model in system RAM is more effective than GGUF quantization. Thus, GGUF formats offer limited benefits in ComfyUI for most users due to sufficient RAM availability.

For running Flux.1 in ComfyUI, use the FP16 text encoder.

(More about ComfyUI settings.)

You can also use FP32 text encoders for optimal results by enabling the --fp32-text-enc argument at startup.

Stable Diffusion WebUI Forge

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

To use the text encoder in FP32 format, launch Stable Diffusion WebUI Forge with the --clip-in-fp32 argument.

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

These example images were generated using FLAN-T5-XXL and T5-XXL v1.1 models in Flux.1. FLAN-T5-XXL delivers more accurate responses to prompts.

Further Comparisons

License

This model is distributed under the Apache 2.0 License.
The uploader claims no ownership or rights over the model.