Show HN: A free text-to-image website that supports SDXL Turbo | Hacker News Reader

Show HN: A free text-to-image website that supports SDXL Turbo

comfyuiweb.com

35 pointsfrankmao20 comments

The homepage of the website allows for free image generation, and in the workflows, there are introductions to nodes related to ComfyUI. In the AI tools section, Pixart-Alpha is currently available for free use. I am still updating this website, and feedback and communication are welcome.

Open articleView on HN

Why you named it ComfyUI? Your name make it sound like you are making that whole interface available online while infact you are only using that as your backend.

How are you able to provide this service for free? What if someone generates tens of thousands of images? Won't that be very expensive for you?

Yes, there is a certain queue for the free service. If too much is generated, the longer the wait time is. The SDXL Turbo model has a manageable cost because it is generated faster.

I assume your plan is to just hope that Stability.AI doesn't bother using you? Clearly this is a commercial service.

https://github.com/Stability-AI/generative-models/blob/main/... doesn't have any restrictions on commercial usage does it?

SDXL != SDXL-turbo, different license

https://huggingface.co/stabilityai/sdxl-turbo/blob/main/LICE...

Huh, TIL: "STABILITY AI NON-COMMERCIAL RESEARCH COMMUNITY LICENSE AGREEMENT"

I learned. I will be replacing the model soon. I haven't looked too closely at the license.

Yeah, I know. I'm not doing anything profitable with the model, but of course if I wanted to commercialize it, I'd apply for a license from stability.ai

Nice! What settings are you using (steps, etc.)?

It's SDXL turbo model. And it only has 1 step.

Okay, and the other settings? Guidance, sampler, etc?

One would think, being described as "Comfyui Online" the expectation is the Comfy UI itself. Since it is not, is this not a misappropriation of the Comfy UI team's identity and their own grass root branding?

Yeah, I am on the fence about flagging this post actually. Using the branding so blatantly feels really slimey.

That would be bad HN etiquette, right? Its not a straight up scam, its seemingly a legitimate service...

Yeah, I assumed it was made by Comfy's author, as maybe a next evolution of their project. I was feeling happy for them.

But since it's not made by Comfy's author, it's definitely not cool to name your project by using their name.

Consider adding a spinner and disabling the generate button once clicked, otherwise there's no indication that clicking the button did anything.

The generated faces are whack

There is spinner on the PC side. You can generate face using https://comfyuiweb.com/Pixart-alpha. The model is perfect and can generate 1024x1024 resolution images.

I also did not see any spinner on the PC side, but cool project and I love how fast it is.

Also see artbot, an actually free SD image generator: https://tinybots.net/artbot

The front end is all in the web app, and the backend generation is done by a swarm of volunteer hosts running on the AI Horde. I donate GPU time on my own 3090/2060 sometimes (albeit for text generation, not imagegen).

https://aihorde.net/

The whole thing is an awesome and tragically underused system.

This is a pretty cool idea.

Recently I have been using SD 1.5 checkpoints with the LCM lora for all of my image generation (generally weighted down to 0.8 strength, with 6 step inference). It's not as fast as sdxl turbo, but at least it does a good job of people's faces.

It would be good to make more efficient use of people's donated resources by making it possible to use LCM lora.

Reply on news.ycombinator.com