/bbwai/ - bbwai

Stable Diffusion, Loras, Chatbots

Mode: Reply [Return] [Go to bottom]

  • Allowed file types: GIF, JPG, PNG, WebM, OGG, ZIP and more
  • Maximum number of files per post: 6
  • Maximum file size per post: 10.00 MB
Captcha no cookies?
Read the rules before you post.
Advanced (canvas, flags, deletion)
Drawing: × size canvas
(For post deletion)
Great for posing and USSBBW right out of the box Ideogram 4
▸
06/06/2026 17:05:05
No. 42534
▸
ComfyUI_00912_
.png

Open file 1.49 MB, 864x1152
ComfyUI_00962_
.png

Open file 3.07 MB, 1536x2048
ComfyUI_00969_
.png

Open file 1.23 MB, 864x1152
Since you can do very precise regional prompting, this allows you to have a lot of control over the poses, and the body shape, which allows you to do stuff which was very hard to prompt for in other models. I've made these images without any loras or control-nets. It sometimes adds a bit too many skin imperfections for my taste, so adding a second pass with Z-Turbo works great to make the skin more idealized.
Anonymous
▸
06/14/2026 12:39:02
No. 42670
▸
>>42534 How are you dealing with the over-aggressive "safety filter" built-in censorship? I've haven't had much time to play with it much, but I've found it's really annoying on simple prompts. Is full JSON prompting more reliable?
Anonymous
▸
06/15/2026 17:19:40
No. 42690
▸
2loras_test__01252_
.png

Open file 1.14 MB, 856x1088
>>42534 ngl these suck. picrel is one I cranked out on my laptop on a local image gen model
Anonymous
▸
06/15/2026 20:39:52
No. 42693
▸
>>42670 Yeah you have to use the json format, the "safety-filter" has only ever activated for me when not using the json format, or having a too small json prompt. I haven't had a single censorship, even with NSFW stuff when using the proper prompting. Try using the KJ Nodes "Ideogram 4 Prompt Builder" node, it makes creating proper json prompts really easy, and allows for really precise control.
Anonymous
▸
06/15/2026 20:46:30
No. 42694
▸
>>42690 Yeah the quality sucks on images 2. and 3., I hadn't figured out the settings yet and was using the wrong scheduler, which was messing it up. The point was to show that, due to the precise control the model allows, you can get very specific body types consistently, which are very hard to prompt for otherwise, without Loras. >>42588 This image, once I figured out the settings, looks really impressive imo. I'm generating the images locally too btw, you can open them in ComfyUI. For regular gooning I'll stick with Z-Turbo since it's faster and I've got a good Lora for it, but if I want something specific, Ideogram 4 is great to have.