Sintetico XL

Recommended Settings:

• Steps: 20 - 45

• CFG: 4 - 8

• Schedulers: I like Uni_PC and DPM++_3M_SDE_Karras

• No need for Refiner or VAE.

What to expect

This model can do good images of cities and buildings, as well as logos and other cartoonish styles. Those are the 3 qualities I wanted this model to excel at, but it can also do other styles and subjects.

It can also generate photorealistic images of people with decent anatomy but not much nudity. I recommend using LoRAs for such tasks as this is not my primary goal.

Sometimes the model may output written text when prompted for it, but it's far from perfect. It's actually random luck to get decent text.

Why I made this

I have a collection of around 40 favorite SDXL models and decided to merge my most used ones. I want a model able to satisfy my personal taste and demand for high quality scenarios, characters for my comics and logos/icons for my projects.

I merged various models and LoRAs. All of which can be found in this collection.

I also used this opportunity to study more of the SDXL space and built a family of models:

1. SinteticoXL (this one): the main model

2. SinteticoXL Turbo: Faster generation, loses quality

3. SinteticoXL Lightning: Even faster generation, good quality

4. SinteticoXL Prude: an experiment to make a SFW version by modifying the base model instead of training on SFW images.

How can you use

Complex prompting is not required, but can be helpful to guide towards your vision.

The model is already merged with FreeU and Self-Attention Guidance. You may try different VAE as per your own aesthetic preferences.

Commercial usage of generated images is allowed.

You are solely responsible for any content that you create using this model. In addition, your use of this model implies that you accept an agreement to not use it to produce harassing, harmful, illegal, or otherwise highly-objectionable imagery.

You may not resell this model or provide it as a service.

Changelog:

• v1.2: Better aesthetics; less models on the merge.

• v1.1 was not published because it didn't showed a significant change in the results.

• v1.0: Updated merged models; Removed FoddaXL from the mix; Fixed add_detail LoRA ratio on merge; overall quality improvement

• beta: Removed DPO from merge. Model merged with FreeU and Self-Attention Guidance and different weights per model. Less saturation than previous version. More creative/artistic results with simple prompting.

• alpha3: improvements with better coherence, better sharpness, better anatomy

• alpha2: removes models with conflicting licenses; renaming for better organization

Roadmap:

I'm experimenting with a way to improve hands and faces. No ETA yet.