Best Performance Setup for FLUX with LoRA on ComfyUI

A TUTORIAL WILL BE COMING FOR BEST WORKFLOWS SUBSCRIBE CHANNEL FOR UPDATES:

https://www.youtube.com/@CSCPRINTSHOPPORTAL

Step 1: Define Your Batch Image Location

Choose your default location for batch image outputs when using ComfyUI. Example folder path:

D:\AI_GENRATION\ComfyUI_WORKING\ComfyUI\custom_nodes\ComfyUI-Inspire-Pack\prompts\cartoon\fluxbatch1.txt

---

#### Step 2: Text File Format for Prompts

Follow this exact format when creating your text file. The positive and negative prompts should be formatted as shown, with the correct number of minuses in between.

**Text Format Example:**

positive: A mysterious sorcerer, cloaked in dark robes, stands in a candlelit chamber filled with ancient scrolls and magical artifacts. His hands glow with swirling energy as he prepares a powerful spell, his face shrouded in shadow. The atmosphere is mystical and ominous. <lora:Cute_3d_Cartoon_Flux:0.7>  

negative:  

----------------

positive: A steampunk-inspired airship flying over a bustling Victorian city. The ship is adorned with brass gears and smokestacks, and the sky is filled with clouds and other flying vessels. The scene is full of adventure and mechanical wonder. <lora:Cute_3d_Cartoon_Flux:0.7>  

negative:  

----

positive: A cute, chubby dragon sits atop a pile of colorful, glowing gemstones in a cozy cave. Its scales shimmer in the light, and its wide eyes are filled with curiosity and mischief. The style is playful and whimsical, capturing the dragon’s endearing charm. <lora:Cute_3d_Cartoon_Flux:0.7>  

negative:

----

#### Step 3: My System Specifications for  These Results

- **GPU:** RTX 4050 6GB VRAM

- **RAM:** 16GB

- **Processor:** Intel i5-13420H

**Performance:**  

For **20 steps** with **Guidance Scale 4**:  

- **With LoRA:** Generation time: 02:25 (7.27 seconds/iteration)  

- **Without LoRA:** Generation time: 02:07 (6.35 seconds/iteration)  

**Resolution:** Use a landscape ratio of 1360x768.  OPTIONAL

---

#### Step 4: GGUF Version for Low VRAM Users

If using the GGUF version of FLUX, this setup works well on lower VRAM systems.

- **My System Specs Using GGUF:**  

 - **GPU:** RTX 4050 6GB VRAM  

 - **RAM:** 16GB  

 - **Processor:** Intel i5-13420H

For **20 steps** with LoRA enabled:  

- **Generation Time:** 02:25 (7.27 seconds/iteration)

---

#### Step 5: Downloads

1. **FLUX Model (FP8):**  

  Download the FLUX UNET model from the following link:  

  https://huggingface.co/Kijai/flux-fp8/tree/main

2. **LoRA Model for Cute 3D Cartoon:**  

  Download the LoRA model here:  

  https://civitai.com/models/677725/cute-3d-cartoon-flux

3. **GGUF Version Downloads:**  

  - **UNET GGUF Model:** https://huggingface.co/city96/FLUX.1-dev-gguf/tree/main  

  - **CLIP T5 GGUF:** https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main  

  - **CLIP L (safetensors):** https://huggingface.co/comfyanonymous/flux_text_encoders/blob/main/clip_l.safetensors

---

#### Step 6: Model Locations

Ensure you place the downloaded models in the correct directories:

- **UNET Model Location:**  

D:\AI_GENRATION\ComfyUI_WORKING\ComfyUI\models\unet

- **LoRA Model Location:**  

 D:\AI_GENRATION\ComfyUI_WORKING\ComfyUI\models\lora

- **GGUF Model Location:**  

 D:\AI_GENRATION\ComfyUI_WORKING\ComfyUI\models\unet

- **CLIP L Location:**  

D:\AI_GENRATION\ComfyUI_WORKING\ComfyUI\models\clip

- **CLIP T5 Location:**  

 D:\AI_GENRATION\ComfyUI_WORKING\ComfyUI\models\clip\t5

Hey, awesome human! If you’ve got an extra coffee lying around (or, you know, the cost of one), I’d love to turn it into creative fuel! ☕️ It’s like magic beans, but instead of a beanstalk, you get more cool content. Thanks for keeping me caffeinated and inspired!

https://buymeacoffee.com/shubzworld