Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

same Blender prompt. same 12 seconds. and Astra’s output looks like it’s in a completely different class. this is why I’m getting more interested in these real software tests than generic benchmark scores. a model can look amazing on paper and still have huge capability gaps inside specific tools.

WesRoth's profile picture

Wes Roth

38,004 subscribers

14,258 Aufrufe • vor 2 Monaten •via X (Twitter)

Wissenschaft & Technologie
LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

Everyone says “Gemini 3 can make games.” So we tested it — with a Metal Slug-style prompt on both Gemini and Gambo. Same input. Completely different output. Here’s the result. #gemini3 #gemini_for_game
0:24

Everyone says “Gemini 3 can make games.” So we tested it — with a Metal Slug-style prompt on both Gemini and Gambo. Same input. Completely different output. Here’s the result. #gemini3 #gemini_for_game

gambo.ai

96,044 Aufrufe • vor 11 Monaten

Image-to-3D tools are leveling up fast. I tested img2threejs by Nick on a Canyon gravel bike from just one image, and the result is already super promising. Still early, but you can clearly see where this is going. Next, I’m going to test the exact same image in Blender and compare the final output. Repo below 👇
0:19

Image-to-3D tools are leveling up fast. I tested img2threejs by Nick on a Canyon gravel bike from just one image, and the result is already super promising. Still early, but you can clearly see where this is going. Next, I’m going to test the exact same image in Blender and compare the final output. Repo below 👇

Givros

47,723 Aufrufe • vor 3 Monaten

Stop relying on generic AI output Train a model on your world, your character, your style, and reuse it forever That is the idea behind LTX Trainer, an open-source training framework that lets you train reusable IC-LoRAs for specific video transformations like colorization, deblur, decompression, reference sheet control, and a lot more Each LoRA gives the model a specific behavior it can repeat across clips, so your videos become more consistent, more controlled, and more connected to your own creative direction Check these amazing examples 👇🏻
0:10

Stop relying on generic AI output Train a model on your world, your character, your style, and reuse it forever That is the idea behind LTX Trainer, an open-source training framework that lets you train reusable IC-LoRAs for specific video transformations like colorization, deblur, decompression, reference sheet control, and a lot more Each LoRA gives the model a specific behavior it can repeat across clips, so your videos become more consistent, more controlled, and more connected to your own creative direction Check these amazing examples 👇🏻

Amira Zairi

28,574 Aufrufe • vor 4 Monaten

We are getting absurdly close to the point where “learning Blender” means learning how to direct an AI. A shot like this looks soft and playful on the surface, but under it is the usual 3D pain: modeling, layout, materials, lighting, atmosphere, animation, and endless tiny fixes until the frame stops looking dead. That is why Kimi K3 matters. With Blender MCP, you can describe a scene like a robotic goat walking through a dreamy field and let the model help build the environment, place the camera, shape the materials, script the motion, and iterate inside the real Blender project. The real shift is not text-to-image. It is text-to-workflow. Kimi K3 does not just give you a pretty output and disappear. It can help move the actual scene from rough setup to something that looks art-directed. Soon the hardest part of 3D will not be the software. It will be whether your imagination is good enough to deserve tools like this.
0:12

We are getting absurdly close to the point where “learning Blender” means learning how to direct an AI. A shot like this looks soft and playful on the surface, but under it is the usual 3D pain: modeling, layout, materials, lighting, atmosphere, animation, and endless tiny fixes until the frame stops looking dead. That is why Kimi K3 matters. With Blender MCP, you can describe a scene like a robotic goat walking through a dreamy field and let the model help build the environment, place the camera, shape the materials, script the motion, and iterate inside the real Blender project. The real shift is not text-to-image. It is text-to-workflow. Kimi K3 does not just give you a pretty output and disappear. It can help move the actual scene from rough setup to something that looks art-directed. Soon the hardest part of 3D will not be the software. It will be whether your imagination is good enough to deserve tools like this.

Rina

45,651 Aufrufe • vor 3 Monaten

First look at the TGL indoor golf arena! Are you more or less hyped? I’m starting to get interested, this looks cool. Justin Thomas and Justin Rose provide the tour. The screen is huge and it looks like a legit sports venue. If they can get some lively fans in there, it could be fun. I’m ready for casual night golf if it’s actually competitive. I see these guys caring, they have equity!
0:19

First look at the TGL indoor golf arena! Are you more or less hyped? I’m starting to get interested, this looks cool. Justin Thomas and Justin Rose provide the tour. The screen is huge and it looks like a legit sports venue. If they can get some lively fans in there, it could be fun. I’m ready for casual night golf if it’s actually competitive. I see these guys caring, they have equity!

Rick Golfs

125,555 Aufrufe • vor 1 Jahr

This AI puts a full production studio in one person’s hands, cinematic-level output. $0.35 per generation. Competitors charge 7× more for the same model. And it’s 70% off. Hollywood is cooked.
0:20

This AI puts a full production studio in one person’s hands, cinematic-level output. $0.35 per generation. Competitors charge 7× more for the same model. And it’s 70% off. Hollywood is cooked.

𝐀𝐍𝐓𝐔𝐍𝐄𝐒

15,381 Aufrufe • vor 7 Monaten

Look at the difference the tool AND the prompt make. Bad generator + lazy prompt: stiff movement, dead eyes, that obvious "AI look" everyone scrolls past. Seedance + a detailed, structured prompt: natural movement, real lighting, real pacing — looks like it came off an actual production shoot. Same product photo. Completely different result. It's not one or the other. You need the right tool AND the right prompt to get agency-level output. Ready to see the exact setup? Check the comments 👇
0:15

Look at the difference the tool AND the prompt make. Bad generator + lazy prompt: stiff movement, dead eyes, that obvious "AI look" everyone scrolls past. Seedance + a detailed, structured prompt: natural movement, real lighting, real pacing — looks like it came off an actual production shoot. Same product photo. Completely different result. It's not one or the other. You need the right tool AND the right prompt to get agency-level output. Ready to see the exact setup? Check the comments 👇

Stijn Feijen

11,987 Aufrufe • vor 4 Monaten

Kimi K2.6 just cooked Claude on design 😭 Same prompt. Completely different output. I asked both to build a cinematic, scroll-driven indie game page. Dark. Editorial. Premium. Here's the exact prompt and what happened 👇
0:17

Kimi K2.6 just cooked Claude on design 😭 Same prompt. Completely different output. I asked both to build a cinematic, scroll-driven indie game page. Dark. Editorial. Premium. Here's the exact prompt and what happened 👇

Muhammad Ayan

162,310 Aufrufe • vor 6 Monaten

this broke my brain. same prompt, two models, one unbroken shot - a carpet becomes a peacock, the peacock becomes a lion. Gemini 4 Argon vs GPT-6 Astra. no edits, no cherry-picking. top is Gemini, bottom is Astra. one continuous transformation, two completely different takes. the real test is the morph itself - does it flow like one living shot, or does it break the second one thing turns into another? I’m not calling it. that’s on you. which transformation goes in the trailer - Gemini’s or Astra’s? drop your pick and tell me why.
0:14

this broke my brain. same prompt, two models, one unbroken shot - a carpet becomes a peacock, the peacock becomes a lion. Gemini 4 Argon vs GPT-6 Astra. no edits, no cherry-picking. top is Gemini, bottom is Astra. one continuous transformation, two completely different takes. the real test is the morph itself - does it flow like one living shot, or does it break the second one thing turns into another? I’m not calling it. that’s on you. which transformation goes in the trailer - Gemini’s or Astra’s? drop your pick and tell me why.

Kinder

10,707 Aufrufe • vor 13 Tagen

I compared MiniMax H3 and Seedance 2.0 on Pollo AI using the exact same prompt, and the results were surprisingly different. The output from MiniMax H3 feels noticeably more realistic, with natural motion, authentic facial expressions, and fewer artificial-looking details. MiniMax H3 shows how prompt refinement and model capability can significantly improve AI-generated video quality. What are your thoughts? Which version do you find more realistic?
0:14

I compared MiniMax H3 and Seedance 2.0 on Pollo AI using the exact same prompt, and the results were surprisingly different. The output from MiniMax H3 feels noticeably more realistic, with natural motion, authentic facial expressions, and fewer artificial-looking details. MiniMax H3 shows how prompt refinement and model capability can significantly improve AI-generated video quality. What are your thoughts? Which version do you find more realistic?

Theo

36,206 Aufrufe • vor 3 Monaten

no this is what a real speedpaint looks like flips, mistakes, zooming in and out, and a CLEAR software user interface
0:16

no this is what a real speedpaint looks like flips, mistakes, zooming in and out, and a CLEAR software user interface

سُمَيَّة

157,990 Aufrufe • vor 2 Monaten

if your app still looks like the one on the left, you're doing it wrong. i used the same prompt. same model. the one on the right used appllama mcp + skills prompt ↓
0:20

if your app still looks like the one on the left, you're doing it wrong. i used the same prompt. same model. the one on the right used appllama mcp + skills prompt ↓

jaimin

259,681 Aufrufe • vor 2 Monaten

Gavin Newsom made it a FELONY for anyone to investigate his rigged elections in California. Imagine 59,000 coin flips all landing the exact same way? This is what tyranny looks like. I’m more interested in what can be done to stop him?
0:11

Gavin Newsom made it a FELONY for anyone to investigate his rigged elections in California. Imagine 59,000 coin flips all landing the exact same way? This is what tyranny looks like. I’m more interested in what can be done to stop him?

Sherri™

30,202 Aufrufe • vor 5 Monaten

Claude Opus 5.5 + Blender is becoming f*cking illegal 😭 A fully mechanical AI generated scarab with articulated legs, metallic armor and moving wings And the difference is actually ridiculous 🟠 Claude Opus 5.5 → $29 / 4 min 🔵 Blender → $16 / 2 min 42 sec Same creature, same mechanical structure, two different workflows The crazy part is that this isn’t just a static 3D model The legs, wings and body all need to move together without breaking the geometry bro we’re getting to the point where building robotic creatures feels more like writing a prompt than doing 3D modeling 💀
0:17

Claude Opus 5.5 + Blender is becoming f*cking illegal 😭 A fully mechanical AI generated scarab with articulated legs, metallic armor and moving wings And the difference is actually ridiculous 🟠 Claude Opus 5.5 → $29 / 4 min 🔵 Blender → $16 / 2 min 42 sec Same creature, same mechanical structure, two different workflows The crazy part is that this isn’t just a static 3D model The legs, wings and body all need to move together without breaking the geometry bro we’re getting to the point where building robotic creatures feels more like writing a prompt than doing 3D modeling 💀

Lummox

11,328 Aufrufe • vor 9 Tagen

GPT Astra vs. GPT-6.1 Sol and every previous Sol generation on the Mario test. Same prompt, same test. Astra still wins, finishing in just 15m 37s for $2.96. GPT-6.1 Sol took 34m 38s, more than twice as long, but cost just $0.64 and delivered the best result of any Sol model. Astra is still ahead, but 6.1 Sol is getting surprisingly close for a fraction of the price.
0:18

GPT Astra vs. GPT-6.1 Sol and every previous Sol generation on the Mario test. Same prompt, same test. Astra still wins, finishing in just 15m 37s for $2.96. GPT-6.1 Sol took 34m 38s, more than twice as long, but cost just $0.64 and delivered the best result of any Sol model. Astra is still ahead, but 6.1 Sol is getting surprisingly close for a fraction of the price.

Luckey Faraday

74,367 Aufrufe • vor 1 Monat

50% cheaper Claude inference with just one line of code change! - Remove → model="claude-opus-4-8" - Add → model="ship-like/claude-opus-4-8" I verified the cost saving in my own terminal by invoking the same Anthropic model with the same prompt. The underlying engineering by Ship is actually interesting, and the patterns can be used in any production LLM stack. Essentially, a trained model is a frozen artifact. Every request performs the same forward-pass, whether it extracts a date or refactors a module, because the compute decision was made at training time, before the request existed. Ship makes that decision at inference time instead. After seeing a request, it searches over executions, involving single models, cascades, ensembles, or harnesses with tools, and serves the cheapest one that will match the reference model's quality. This is not a basic router, because picking a cheaper model per query doesn't ensure the cheaper model preserves the original's behavior, like output shape, tool-call patterns, and refusals. Ship measures this equivalence directly. Outputs stay distributionally indistinguishable from the reference model, not token-identical, since two calls to the same model already differ, but they are indistinguishable in capability and behavior. Of course, some requests execute cheaply and some cost Ship more than the customer pays, but the price per request is still a flat 50% off either way, so the execution-cost variance moves off the application's bill entirely. The video below depicts the cost savings and output in my real invocation, and I partnered with the team to put this together.
0:12

50% cheaper Claude inference with just one line of code change! - Remove → model="claude-opus-4-8" - Add → model="ship-like/claude-opus-4-8" I verified the cost saving in my own terminal by invoking the same Anthropic model with the same prompt. The underlying engineering by Ship is actually interesting, and the patterns can be used in any production LLM stack. Essentially, a trained model is a frozen artifact. Every request performs the same forward-pass, whether it extracts a date or refactors a module, because the compute decision was made at training time, before the request existed. Ship makes that decision at inference time instead. After seeing a request, it searches over executions, involving single models, cascades, ensembles, or harnesses with tools, and serves the cheapest one that will match the reference model's quality. This is not a basic router, because picking a cheaper model per query doesn't ensure the cheaper model preserves the original's behavior, like output shape, tool-call patterns, and refusals. Ship measures this equivalence directly. Outputs stay distributionally indistinguishable from the reference model, not token-identical, since two calls to the same model already differ, but they are indistinguishable in capability and behavior. Of course, some requests execute cheaply and some cost Ship more than the customer pays, but the price per request is still a flat 50% off either way, so the execution-cost variance moves off the application's bill entirely. The video below depicts the cost savings and output in my real invocation, and I partnered with the team to put this together.

Akshay 🚀

64,482 Aufrufe • vor 3 Monaten

Garnacho vs Cunha in 16 seconds....same scenario,same kits colors ,completely different outcome.
0:16

Garnacho vs Cunha in 16 seconds....same scenario,same kits colors ,completely different outcome.

Ski

943,651 Aufrufe • vor 1 Jahr

real estate has always been a frozen asset. mey real unfreezes it. your property token is tradeable onchain, and it sits in your wallet right next to your $USDC yield. same asset class. completely different mobility.
0:10

real estate has always been a frozen asset. mey real unfreezes it. your property token is tradeable onchain, and it sits in your wallet right next to your $USDC yield. same asset class. completely different mobility.

mey real

16,390 Aufrufe • vor 7 Monaten