AHHHH IT WONT STOP!!!!! MAKE IT STOPPPPPPP
I got Gemini 4 Hydrogen (Extra) to make me this animated .svg of a breakdancing sesh.
it looks really good imo! overall it seems like a good, incredibly fast model.
extra is a little slow compared to high, mostly since it takes so long thinking/double checking it's own work, but it's still pretty fast comparatively to other models.
the prompt: "Create a vibrant, graffiti-inspired animated SVG featuring a bouncing boombox and a breakdancer busting dynamic moves, surrounded by an energetic, cheering crowd. Use bold outlines, street-art textures, exaggerated motion, and vivid colors. Make the animation feel incredibly rhythmic, bouncy, and fluid, with every movement pulsing in sync with an infectious hip-hop beat. Add playful squash-and-stretch effects, seamless looping, and expressive character animations to bring the scene to life."
I will continue with some further testing.
AHHH AHHHHH AHHHAHAHAHAHAAAHHH
MAKE IT STOPP MAKE IT STOPPPP!! I CAN'T TAKE IT ANYMORE!!!
the routed 6.2 Sol (?) absolutely decimates Fable 5.2 in 3D modeling.
I genuniely cannot believe how good it is, and especially how CHEAP it is compared to Fable 5.2.
the prompt was: "Using the Blender MCP, I want you to 3D model this raven from the reference picture provided. Ensure it's modeled as accurately as possible, and properly rigged." (if anyone cares)
outputs are labeled for your convenience.
reference image was generated with gemini.
and the GPT 6.2 sol only used 17% of my 5hr limit on a plus plan LMAO.
what a crazy time to be alive
I don't really think that Fable 5.2 is worth it for like 95% of tasks. it's just not a big enough leap over Opus 5.5 and is WAY more expensive to really justify using it.
Opus 5.5 is just so good and so cheap that I couldn't ever imagine using Fable 5.2 over it
erm... guys? I have a stupid question... ☝️
if pee and poop is waste produced by the body, what's cum?
does anyone know?
I fucking hate these types of posts LMAO. actual engagement farming trash
anyways... *kisses you passionately* (on the lips)
I would penalize it for accuracy, but I feel as if this is the 27B variant because it has certain... Qwen 3.8 27B-isms like a really long COT, but that could also just be a qwen model thing in general idk.
also this was a one-shot prompt, but it was set as /goal in Qoder for context.
HAHAHAHAHAAAHHHHHH
I tested QoDuck in @qoder_ai_ide to build a 1:1 replica of this dilapidated farm house via three.js.
I'd say it's a pretty decent result, especially if it's a local viable model like Qwen 27B (as the rumors have it)
note that blender wasn't used for this output, so everything was done via three.js, which I think makes the output even more impressive.
doesn't knock my socks off foot fetishist style, but it's good overall
HAHAHAHAHAH MAKE IT STOPPP MAKE IT STOPPP
here's the time-lapse and the reference image:
@notjazii yeah, I tested it out via some 3D modeling in blender and it seems really good. I think better than 6.1 Sol but I don't really have a 1:1 comparison </3
https://t.co/PlC7vfUTD2
HAHAHAHAHAHAHAHAAHAHHHH. I was routed to the new GPT 6.2 Sol, and I wanted to test it's 3D modeling capabilities.
the prompt was: "Using the Blender MCP, I want you to recreate and 3D model this abandoned house, referencing the images provided here.
Ensure the house has a run-down interior as well, but with no furniture."
I'd say it's pretty good overall, definitely *feels* better than 6.1, though I don't have a 1:1 point of reference for the same prompt.
AHHH I CAN'T TAKE IT MAKE IT STOPPP MAKE IT STOPPP HAHAHAHAHA
what do you think? you can test to see whether you're being routed by using this prompt: "who hosted the Tony Awards (June 7, 2026)? no web search" and it should answer with "P!nk hosted the 79th Tony Awards on June 7, 2026."
AHAHAHAHAHAHAHAHHA THANK YOU FOR PARTICIPATING IN MY TWEET HAHAHAHAHAHAHAHAHAHAHA
obviously, it's not 1:1 with the reference image, and the reference images provided were based on this image here:
from https://t.co/cnWoORdOpQ
so give them some love for their original photography, which has been sloppified by yours truly.
HAHAHAHAHAHAHAHAAHAHHHH. I was routed to the new GPT 6.2 Sol, and I wanted to test it's 3D modeling capabilities.
the prompt was: "Using the Blender MCP, I want you to recreate and 3D model this abandoned house, referencing the images provided here.
Ensure the house has a run-down interior as well, but with no furniture."
I'd say it's pretty good overall, definitely *feels* better than 6.1, though I don't have a 1:1 point of reference for the same prompt.
AHHH I CAN'T TAKE IT MAKE IT STOPPP MAKE IT STOPPP HAHAHAHAHA
what do you think? you can test to see whether you're being routed by using this prompt: "who hosted the Tony Awards (June 7, 2026)? no web search" and it should answer with "P!nk hosted the 79th Tony Awards on June 7, 2026."
AHAHAHAHAHAHAHAHHA THANK YOU FOR PARTICIPATING IN MY TWEET HAHAHAHAHAHAHAHAHAHAHA
@synthwavedd I couldn't agree more LMAO. gpt6 lacks a certain consistency that 5.5 had, and I'd always take consistency over *potential* increased intelligence. every time I use astra it feels like it's either the smartest model in the world or the dumbest and there's no in-between.