Ultraforge initial beta findings

Per request, here is a new thread regarding bugs/issues related to the initial Ultraforge beta release. Please feel free to add/echo any findings as you can.

- no obvious logging of exceptions could be found
- shuts down within 3 seconds of starting
- crash on 2nd image generation regardless of text used
- crashes more likely with longer prompts, specifically when adding "seamless repeat pattern"
- only able to render the default image
- potential workarounds for subsequent generation crashing:
-- saving project as '.ufo' file before every generation helped me
-- someone else mentioned deselecting nodes before generation helped them, but I have yet to test this

So far, awful. I've been using Bing AI Image Generator while I waited, with a detailed prompt. Some weird details - but overall beautiful images, very usable if you have to run it a few times to get what you want.
UF crashes almost every time. Sometimes on the 1st attempt after starting the program. Tried putting in just one person - girl in black leather armor. Looks like a bad Daz3D person from 10 years ago. Head cut off, no details, not usable for anything.
Running 10th Gen Core i7 w/64 GB RAM, software installed on an M.2 drive, RTX 3090, Win 10.

Hello,

I am new to this forum but have been using Filter Forge since version 9; I came here to find more guides about how to use Ultraforge. I downloaded the Beta today.

It works perfectly for me, but I have not used other AI tools. I have Ryzen 5900X, 64GB RAM, and RTX 4070 Ti on Windows 11 Pro 22H2.

I created an image using the prompt: hippo sitting in a chair playing a guitar, wood fireplace, dark lighting, super detailed
I hope the attachment button works below... first post and all
The seed was quite high, but even low seeds result in funny results.

To the developer, thank you for creating this software and I hope I can learn more about how to use it soon.

Requirements check passed, but I can't render anything!!! (win11, i9, rtx3080Ti)

The output image with the default 'kitten' is just a square with 'concrete texture'.
If I change the texts, the result is another variation of the concrete textured square.
If I set the STEPS value to 1, I have the expected multicolored square.
If I change the value to 30, I have the concrete textured square, the rendering time is much longer, but the 'render strip' is always stuck at 5%.

Installation, I changed the default drive C: to D: but I see that the main files (about 2GB) are still on C: under Users/Public/Documents. You should let the users decide where to put the main files!
For testing purposes I reinstalled everything on C: but the results are the same, no expected output image.

Quote
Ricco Biggo wrote: I created an image using the prompt: hippo sitting in a chair playing a guitar, wood fireplace, dark lighting, super detailed

Hello,
Can you share a screenshot of your nodes for testing purposes?
Thanks

Quote
STEPH08 wrote: Interesting, I need to check

That happens if you have the Ransomware Protection on (I do), so a new program can't access important folders inside userprofile, program files, etc.

Quote
Grimmy wrote: How can I get a refund?

Drop a line to https://filterforge.com/contact/ -- they'll issue a refund, no questions asked. Include your order details into the request.

Quote
cadranni wrote: Tried putting in just one person - girl in black leather armor. Looks like a bad Daz3D person from 10 years ago. Head cut off, no details, not usable for anything.

Quote
Grimmy wrote: The first prompt I tried was an awful mess. It rendered some shapes that were complete nonsense and nothing to do with what I asked for.

Quote
Grimmy wrote: When comparing this to Microsoft Bing's AI image generator this is complete trash. (Bing generates 4 images in the same render time that are actually useable)

Ultraforge's generative AI nodes are based on Stable Diffusion, which is not very good out of the box. It requires some tweaking to get good results, and you can download and plug external checkpoints which are much better.

Here's a good Stable Diffusion prompting guide which shows that even the base SD 1.5 model can produce good results: https://stable-diffusion-art.com/prompt-guide/

And if you want even better results, grab a checkpoint file (.safetensors) at https://civitai.com/models, and put it into your Ultraforge vault folder (C:\Ultraforge\SD 1.5 Extras\Checkpoints). The pic below shows how to search for checkpoints that are compatible with SD 1.5:

Quote
ALZHEM wrote: we will be able to use different models to achieve different results

You already can use custom models -- see my post above for a quick guide.

Quote
ALZHEM wrote: I saw! That's amazing, thanks! Are you planning to add support for SDXL soon?

Absolutely -- as soon as we fix the bugs, that is. SDXL will exist side by side with SD 1.5. Also, we plan to support LoRAs (and possibly LyCORISes) for SD 1.5 -- there are many fantastic LoRAs available, and we want them to be usable in UF.

And there's the ultimate Holy Grail, ControlNet, but that will take a while.

Quote
Vladimir Golovin wrote: Absolutely -- as soon as we fix the bugs, that is. SDXL will exist side by side with SD 1.5. Also, we plan to support LoRAs (and possibly LyCORISes) for SD 1.5 -- there are many fantastic LoRAs available, so we want them to be usable in UF. And there's the ultimate Holy Grail, ControlNet, but that will take a while.

CAN'T WAIT!!!
A last thing before I go to bed, are you also planning on adding the PNG metadata workflow that ComfyUI has? I love that a PNG can hold the nodes of how it was created.
Thanks for all the answers!

Quote
STEPH08 wrote: Requirements check passed, but I can't render anything!!! (win11, i9, rtx3080Ti) The output image with the default 'kitten' is just a square with 'concrete texture'.

fixed !! I have a hybrid system, for unknown reasons the app was not using the graphics acceleration of the Nvidia card. I manually changed the nvdia setting for the app from Auto Selection to High Performance.