ATTENTION Currently, the only way to use TinyBreaker is through ComfyUI by installing:
these nodes can be easily installed using ComfyUI Manager.
Workflows
The example images I've posted here on CivitAI include embedded workflows for use in ComfyUI. First, click on the image to enlarge it, then drag and drop it directly from your browser window into ComfyUI to automatically load the associated workflow.
However, if that doesn't work, base workflows are available here:
TinyBreaker Model
TinyBreaker is specifically designed to consume minimal resources while still achieving stunning visuals with little effort. It's a hybrid model that utilizes PixArt as its base and Photon as a refiner. At present, the project is entirely experimental.
Overview
Resource-Efficient: Designed for minimal resource usage, likely less than 8GB
Generation Speed: Images at 1536×1024 in ~12 seconds on an RTX 3080.
High Prompt Adherence: Generated images closely follow the given prompt.
Hybrid Model: Combines PixArt and Photon models.
Experimental: Currently in the Prototype0 phase, fully in testing.
Description
This version is required to use the "Tiny Upscaler" included in the v0.3.0 of the custom ComfyUI nodes. The previous version (prototype0) excluded certain VAE components to save space. This update embeds all those components into the .safetensors file, enabling the upscaler to function properly and paving the way for potential inpainting features in future updates.
FAQ
Comments (12)
is there a way to make this work in Forge
I think this is a wonderful upgrade.
It seems that artifacts and distortions in the details have been reduced compared to previous versions, and the newly added upscaling feature is very convenient.
Many of the things I was secretly hoping would be improved have indeed been refined, making the workflow more user-friendly and easier to work with.
I also truly appreciate how carefully the development is handled—every change is clearly documented, which greatly helps in understanding the workflow more deeply and is incredibly helpful from a user’s perspective.
This will not work on linux, it can't find the "checkpoints folder and I can't get the node to find it.
Everything works perfectly, even implemented a groq method of refining the prompt based on pixart's params.. Fun stuff
The samples are shockingly good, this looks like flux / midjourney quality on a 3GB model
Is it possible to train your LORA using this model? What is the best way to do this?
Hello, today i update ComfyUI, now TinyBreaker stop works with this error
get_t5_encoder_custom_class.<locals>.T5EncoderModelCmfy.forward() got an unexpected keyword argument 'embeds_info'
Someone can help please??
Resolved by updating ComfyUI
No updates?
Soon! I'm putting together something like a "prototype2" with a new workflow for more varied and striking style generation. (some previews: https://civitai.com/posts/23947060).
To be honest, I'm deeply focused on another project at the moment, leaving limited time to give TinyBreaker all the love it deserves.
Thanks for sticking with this wild experiment!
@Photographer Nice