#freestyle Homophones, Multisyllabic double entendre, Misdirection rhyme, #sonicwriting
07.03.2026 16:30
π 0
π 0
π¬ 0
π 0
#freestyle Homophones, Multisyllabic double entendre, Misdirection rhyme, #sonicwriting
United with fam, the hand that plans on our arrival
Some said don't waste your time yo that mission's suicidal
Saw vision deep when the lights is on pondering survival
Rivals face to face to lace the taste of wasting time in cycle
'Cause I was told the bones of old injustices
Have a way of rising...
.γ Λ γ. γ β§ γ Λ γ. γ β§ γ Λ γ .γ β§ γ. γ Λ γ.
If you wish to participate in our upcoming exhibit in The Netherlands for Alzheimerβs, the call for artists is now open.
To register, please visit postcardartexhibit.com
Also visit our FAQ page for any questions: postcardartexhibit.com/faq
We hope that many of you can join us πβ€οΈ
Pinocchio (1940)
#animationclips #sakuga #animation #2danimation
LoRA makes "intruder dimensions" in singular vectors that differ from the pre-trained model while full fine-tune remains spectrally ~ to the base model. The takeaway LoRA excels at instruction fine-tuning with smaller datasets, while full fine-tuning shines in continued pre-training scenarios.
Graphic with a dark purple background. In the centre it has a drawing of a red awareness ribbon. It says "World AIDS Day 2025". The Equality Network and Scottish Trans logos are in the bottom right corner.
Today is World AIDS day β€οΈ
World AIDS day has been marked every year since 1988, to:
β€οΈ Fight the stigma of HIV/AIDS
β€οΈ Show solidarity with those living with HIV/AIDS & honour those we have lost to AIDS-related illnesses
β€οΈ Raise awareness of the testing, treatments & interventions available
"My name is Alexander, King of Macedonia. I will grant you any request."
"Move over, you're blocking my sun."
Approximately 7% of American men think they can beat a grizzly bear in an unarmed fight, according to a 2021 YouGov survey.
2025 internet in a nutshell - trust me bro just one more merger, if it ain't a monopoly then it's fine bro
github.com/HKUDS/LightRAG
World Labs has released Marble, a commercial world model that turns text, images, videos, and rough 3D layouts into persistent, editable 3D environments. The company is led by Fei-Fei Li, the Stanford professor known for creating ImageNet, the dataset that taught computers to see.
Kimi K2 - ~5 million
Gpt-4: estimated to need $100 mln for training
Some later antropic model - took like 20-30 millions
Deepseek v3 - 8 millions or so
Spatial Intelligence: Photogrammetry, neural radiance fields (NeRFs), and 3D Gaussian splatting for digitizing reality.
Visual Intelligence: AI-powered pose estimation, segmentation, depth inference, and relighting.
... and more
So βpromptingβ at a high level is equal parts art direction, data engineering, and patience. Itβs not typing fancy words into a box itβs designing and controlling a generative system down to the smallest variable.[7/7]
Doing it ethically makes things harder. Most artists refuse dataset licensing even when offered payment, so finding or creating legitimate material often becomes its own long, frustrating research project.[6/7]
Thatβs just for one or two characters. If the right LoRA doesnβt exist, you build one yourself: collect and clean a dataset, tag it properly, train the model, adjust learning rates, and hope it doesnβt overfit or collapse.[5/7]
Once it renders, youβre nowhere near done. You inpaint bad spots, fix hands, correct lighting, adjust anatomy, balance colors, and sometimes re-render specific regions just to clean up the smallest inconsistencies.[4/7]
Then comes structure control. ControlNet or T2I adapters for pose, depth, and composition. You feed it skeletons, edge maps, or references to make sure your characters donβt drift off or grow an extra arm mid-render.[3/7]
You write a base prompt and a negative prompt, select LoRAs or LyCORIS that match your style, add tags, balance their weights, fine-tune descriptions, and test different combinations until the model starts doing what you actually want. [2/7]
AI βpromptingβ is way more than just typing words. If you want consistent, high-quality images, itβs a full creative pipeline involving composition control, model tuning, and constant iteration not a magic spell or a lucky roll that most believe it is.[1/7]
For those who need to hear it today π³οΈβπ
