Floopers966 (@floopers966) profile picture

He’s small but I wouldn’t call him punny.

Joined October 2021
Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)Photo by Floopers966 (@floopers966)
Some seedance 2.5 with a glm 5.3 flash created prompt and qwen 2.1 reference image.
1
1
120
The Denny's where Nvidia was born has removed the sign honoring the company nbcnews.com/tech/tech-news/d…
59
498
29
4,830
223,708
Image model comparison! GPT 2.6 / Seedream 5.5 / Flux 3 Image / Krea 3 high - more in reply
1
95
and now Krea 2 (local) / Qwen 2.1 (local) / Moss SFT (local) / Minimax H3 Image (local)
146
Seriously impressed with MiniMax H3-Image. Far higher resolution than any other model and allowed for reference inputs.
104
Apparently Trump had a falling out with this lady.
24
At some point in our future, we'll get to watch a deorbiting data center as the thrusters are disabled and it succumbs to the nvidia geforces.
16
Yay OpenAI, GPT Image 2.6 looks to have finally fixed the weird moire patterns on most generations. Everything I'm getting out of 2.6 is super clean.
🤖 Made with AI
89
For those running Deepseek 4.1 Flash 1017 abliterated mlx locally, many versions come with a quantized ngram that's SSD offloaded. One can download the full nonabliterated model also and just have Opus modify the fp4 quantized ngram back to full fp8 so you don't lose that quality with little speed difference.
1
63
Changing my prompt style back to forcing closeups for anyone talking. Seems to work better with Minimax H3 and Seedance 2.5. I've been seeing all the X.com greats do it and never thought to modify my own instruction.
1
162
Every time a new text to image model comes out. #Flux3 #MOSSImage
1
259
Flux 3 Klein can be impressively good at some things (second pic with giant) but then fall down really hard on other content. Moss (at the same model size) is far more evenly capable.
1
100
I'm here for the 100s of Tesla Roadster videos of the thrusters going haywire.
45
With the new Flux 3 Klein, they're recommending cfg 3.5 for base, but I'm finding far more coherence with cfg 5.
1
72
Getting pretty reliable stuff out of it now with cfg 5 / res_2s / simple / 25 steps
38
Flux 3 image on api is seriously impressive. It's also a lot more coherent than gpt image 2.5.
1
69
Unfortunately the default ComfyUI workflow template is using CFG 1, when it should be at 3.5. The grain falls away and the colors really pop.
84
The new Qwen 2.1 image model's reference input image game is strong. It's seriously impressive given how small it is.
295