not a coinkydink he bought Huggingface relatively right before this btw, that's (probably, hopefully not but you know) going to end up becoming an even worse thing
7
honestly it feels like the image gen needle hasn't moved much if at all, Krea 2 still the best overall especially with ControlNet, Anima for live vibedrawing, though Moss is also getting there, Klein 2 9B and H3 with P-H3 for edits/refs, maybe Qwen 2.1 for transparency/bg removal
1
79
i'm almost convinced most AI labs don't test on PC screens and only target phone screens and just go "eh normies and zoomers don't actually zoom in or even own a computer so they won't notice when stuff is melting even at 8K anyway"
15
H3 Image is alright, but honestly P-H3 is also still good and I can also still use the base H3 model it's applied on top for video, also H3 Image even at 50 fucking MP stuff still looks like it's melting so what even and how even, I've seen sharper more defined stuff at just 4
1
43
relatability through simplicity is universal because it appeals to both beginners "i can make this too!" and pros "i can make this, but better!", but also no one fetishises uglyness and unfinishedness more than retarded leftoids and that's what anti-AI luddites are
104
never got into LLMs because they're kind of a pain local tbh, compared to the relatively ezpz setup, params and usage for image/video/audio, though i'm sure it's the opposite for some/many, man, i got the audiovisual 'tism not the coding one, maybe Strata will be the one ™️
36
it's cool AI lets people make things, and it's (not entirely) dumb to go "presentation matters more", but i see these "show your AI thing" posts & 99% of it looks the same and i can't tell whomst'ved made what, still only about 100 people using AI in the world have a unique style
27
augmenting "trad" techniques with AI has always been far more interesting and cooler to me than rolling the noise roulette, though also seeing the math gacha you get is interesting on its own, but this just gives you much more control objectively saving time too
1
83
i AM grateful every time one of my pics gets at least one like, and also lots of likes which is rare so i still dunno how it happens, though it must be Grok's image recognition and matching that to existing trends even without tags or post text (o algo, literally)
4
49
this is the best H3 Edit method now, no reason to use any of the older edit nodes and workflows anymore tbh, also it's a lora + smaller text encoder + VAE, which means it works on regular H3 checkpoints, including FastH3, also works with the H3 Anime VAE civitai.red/models/3025416/m…
Today, we open-source P-H3-Image, an image generation and editing model built from MiniMax-H3 by @MiniMax_AI. A SOTA video model, turned into an efficient open image model.
- 𝗛𝗼𝘄 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗶𝘀 𝗶𝘁? Generate or edit images 5× faster, in 8 steps instead of 50. A 4 MP image takes 2.6 s instead of 12.4 s. The model is 7× smaller than H3, so high-resolution images are a light download and run on a much smaller GPU.
- 𝗛𝗼𝘄 𝗴𝗼𝗼𝗱 𝗶𝘀 𝗶𝘁? Strongest at photorealism, documentary and editorial photography, film stills, and graphic styles with short quoted text. It renders 2 to 16 MP and edits with up to 8 input images.
-𝗛𝗼𝘄 𝗱𝗼𝗲𝘀 𝗶𝘁 𝘄𝗼𝗿𝗸? We reworked every component of H3: the prompt upsampler, the text encoder, the diffusion schedule, the transformer, and the decoder. Use it with Diffusers or ComfyUI.
📄 Tech Report: pruna.ai/blog/oss-p-h3-image…
🤗 @huggingface Open Weights: huggingface.co/PrunaAI/P-H3-…
👉 @ComfyUI Native Support: huggingface.co/PrunaAI/P-H3-…
1
7
251
only when that job is just means to money/obligation with nothing else to it tbh, it's still fun to sketch/draw and make stuff, and even more fun with AI assist, it's also OBJECTIVELY (don't@me) less fun to just press generate and go "i maded this" without any other involvement
I want AI to do my job better than me.
Not "assist" me. Not "augment" me. Not be my "copilot."
Replace me.
Every hour I spend doing something a machine could do is an hour I'm not spending with people I love, building things I care about, or just doing nothing at all.
People treat "AI can do your job" like a threat.
To me it sounds like a release date.
1
126
static vs dynamic shot, depends on intent & preference, i enjoy thicker sketchy lines more so right, but it's also dumb that cleaner detailed images get insta associated with "AI LE BAD" & sketchy scribbly SOVL with "HUMAN LE GOOD", even though i also do the latter on purpose
1
64
it's still so funny that no one treats "AI" like a meme sentient robot being more than luddites, especially image AI that doesn't even use their LLMs as an actual one and "thinks" the least if not at all, true gigachad nerdheads know it's just another computer program
1
80
as much as i liked Moss and its variety i ended up still sticking to Anima+Krea 2 lol, Krea 2 is still better at styles especially with stuff like link related, and has the best ControlNet so far, who ever knows how long, still hoping Krea 3 will be good
docs.google.com/spreadsheets…
1
3
164
sketch discourse is always retarded, a sketch is 1) the fastest "snapshot" of your brain into a medium 2) not yet finished, amateurs and dogmacucks treat it like a holy building foundation, true chads don't even need a base sketch, because the entire piece is one until it's done
Speedpaint trying to prove is not traced: squiggly/hairy lines (VERY begginer mistake)
unconfident lines
no anatomy fundamentals/no proportions
Sketch made MONTHS AGO: perfect anatomy and proportions
Im not even gonna address the generated parts in other designs
3
286
now that models won't be trained on Reddit anymore (why even in the first place too), AGI will finally be achieved
Reddit is getting rid of logged out browsing. They want to require everyone to be logged in to access the site. They’re claiming it’s to protect against scraping which is hilarious because the site has already been scraped. End of an era. No more googling “question + Reddit”
52
daily can't wait for the meme corpo bubble to pop specially his
OpenAI CEO Sam Altman says it’s “quite terrible” that people have so much DRAM sitting in their own PCs instead of being used in data centers.
“Everyone have a bunch of the world’s limited supply of DRAM sitting in their own computer. And be used 5% of the time,”
He argues that cloud data centers should make better use of limited memory supplies instead of people running AI models locally.
1
38