SUPERMACHINE Reviews

Dear Supermachine Team,

I would like to share some thoughts about my experience with the platform. Currently, I am trying to generate images with Supermachine, and unfortunately, I have to say that the process often takes a very long time. I find myself staring at the screen, and after 10 minutes, the page is still not fully loaded, and I have also been waiting for the image for the same amount of time. Admittedly, there are times, like in the middle of the night, when it is faster, but overall, I find the speed to be inadequate.

In the new interface, I have the option to increase the steps up to 80, but I do not see a sampler, which only costs credits unnecessarily without providing any real benefit.

Now we are at the 15-minute mark, and I can only see half an image in the preview from SD 3.5 Large. I sincerely hope that the situation improves in the future. What good are 4000 credits to me when I have to frustratedly switch to Replicate and pay their high prices?

I just saw that the three images were finally generated after 18 minutes. That’s it for me today; I will log out now.

I truly expected more from the platform.

While having a lot of models, etc. Supermachine overwhelms you with all the models rather than assists you. Superclone is their way of trying to keep characters consistent using one image, but it falls VERY VERY short. If there were a way to provide more images for consistent characters, it would be better.

For the moment, I will give it 2 tacos, just because you can spam image generation, whether it would be helpful for you or not. I doubt it. I've spent more credits deleting photos and batches and recreating stuff than actually having useful images.

Prompt assistance is absolute rubbish; please do not use it. It gives you a verbal diarrhea of words and stuff it with more things. Hindering your results. As of this moment, I do not have a use case for it, unfortunately. Other default image gen tools do it much better and quicker, hopefully the devs can bring Supermachine on the right path, perhaps help us create our own loras so superclone actually works better. Perhaps superclone works for animated characters than human faces. Quality seems to have dropped off in the last 2 weeks or so. Will continue to wait for the development of this product to see where it goes. So far.... not great.

I bought the first tier a few month ago. It has a learning curve, but after i got into it, upgraded ti tier 3. so many great AI models, so many possibilities .

Big ideas, poor implementation.

1. Stuart keeps adding new features with a lot of hype, then abandons them. I expect some consistent game plan from a paid product. Also, with less and less customizable options, I don't see the appeal for power users anymore.

2. My main issue is how slow the dashboard is. I can't sit and wait for 5 minutes to generate one image. Honestly, SM would be better as another wrapper for enterprise APIs than trying to run heavy image models on consumer-grade servers.

3. Apart from soul-draining generation speed, image quality is also handicapped compared to original models they rebranded (SD, Flux, HiDream). Clearly, they had to be quantized to run on a consumer GPU.

The new UI looks promising, we'll see. I wish Supermachine to succeed, I just felt like I need to share my experience so far. And so far, my time and money could be spent better on something else.

I would get more use out of it if I could plug in my own Runware or Fal API keys. Actually, it might be necessary for SM to add BYOK to survive on this competitive market - the images I get from Runware are much better, blazing fast, and cost less than 1/2 cent each.

Yes, it's a lifetime deal, but it's impossible to generate 1000 images per month when it's so slow.

Hope SM will improve in the future,
Cheers!

Like some of the others here, expectations were high. This probably should have been my first red flag, especially with AI. Generally, it takes a lot of credits to get to a small batch of usable images. However, with Supermachine, it appears you would not even get a batch of usable images. After playing with it for a couple of hours, I didn't render one usable image. I was nearing 100 credits when I just called it a wrap. I wasn't looking for perfection. I was trying to test the different variations, like the product videos suggested. There was no efficiency in running prompt variations, sliders, model changes, etc., for 2 hrs to not get one usable photo. I even tried using one of the "consistent characters" just to get an image. It appears as if this product might fall in line with some of the software you see in markets such as JVZoo. I used one-sentence prompts. I elaborated and used detailed prompts (scene, pose, lighting, camera, mood, and background), and nothing worked to generate even a reasonable image.