TIP posted:

mobby_6kl posted:

KakerMix posted:

KwegiboHB posted:

spinning meat
Steps: 48, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 4, Size: 512x512, Model hash: 99fd5c4b6f, Model: seekArtMEGA_mega20, Denoising strength: 0.69, Clip skip: 2, Version: v1.2.1, Hires upscale: 1.5, Hires steps: 48, Hires upscaler: Latent (nearest-exact)

VectorSigma posted:

pixaal posted:

Different models have different strengths. Even with bing though keep trying and rewording. You can usually find something even if you have to get creative.

You can make planets in most things with "fantasy hand drawn map wrapped around a sphere floating in space" generic planet will mostly just get Earth clones.

ymgve posted:

possible PTSD from watching Norwegian 80s children shows?

pixaal posted:

Honestly I used 1.5 and 2.1 for a very long time like until March I was just using them with control net and other plugins, get to know the base model and how it's worded most other models are just trained on to of 1.5, so learning to make good images with base 1.5 is a good starting point.

KwegiboHB posted:

I'm not sure if there's a widely accepted 'general model' so here are a bunch.
These are some older models I can personally vouch for. I know you can get quality out of these.

And finally for MUCH further reading, https://sdtools.org/. This giant wheel has doubled in size since I first linked it and I'm personally not even half-way through some of these. Still haven't even started using ControlNet lol. I say it's ok to feel lost in all this, just have fun with it and see what you can come up with. Be sure to share with the class!

These are excellent monster trucks!

AndrewP posted:

I love kittenfrog

hydroceramics posted:

pixaal posted:

RIP Syndrome posted:

Great. I'll be posting a lot less soon (work), making you the designated spam-slack-picker-upper.

hydroceramics posted:

A breakfast food known as a Roman Glacier;

pixaal posted:

Really large collections like this look good as timg

I miss the AI that understood cup sizes and you could say man with DDDDDD tongue and get horrors of a tongue swollen to hell (it looked like a bloated corpse but someone struggling for air at 128x128) SD with the concept of sizing dynamically would be nice.

KakerMix posted:

These systems have a language they speak that isn't like how you and I speak to each other, so asking for 'REALLY BIG HAND' doesn't mean anything to it because it is only going to summon images for tags it has and doesn't understand language well like that if the tags don't match. Asking for something like 'a dude in a blue hat' works because presumably there are lots of images with those tags, but 'big hands man' not so much. To get what you want you'd really want to inpaint to side-step context and just get a big hand. Or, like the model linked above, use user-created models to get whatever you want because boy there is a lot. Note that most trained models right now are SD 1.5, which is not XL which means it's 512x512 and has minor difference in how it prompts.

EDIT

pixaal posted:

If I end the batch in CMD with control+C it always recovers to no queue, maybe reboot your computer? Restart, shutdown may cause windows to suspend instead. A1111 worked the same, though A1111 sometimes didn't release VRAM in early builds and needed a reboot or you'd have no VRAM.

ComfyUI Github FAQ posted:

You can use () to change emphasis of a word or phrase like: (good code:1.2) or (bad code:0.8). The default emphasis for () is 1.1. To use () characters in your actual prompt escape them like \( or \).

You can use {day|night}, for wildcard/dynamic prompts. With this syntax "{wild|card|test}" will be randomly replaced by either "wild", "card" or "test" by the frontend every time you queue the prompt. To use {} characters in your actual prompt escape them like: \{ or \}.

Dynamic prompts also support C-style comments, like // comment or /* comment */.

Hadlock posted: