So, the new Gemini-3-Pro-Preview, a few points:
- The level of details is like nothing we've seen before; for some shots it doesn't look like AI at all. After the initial generation, the model can upscale the image without losing details to 2k or 4k, depending on the settings.
- It is not very stable yet; in fact, you can only try it through the API, and not all generations are successful there. I was unable to generate an image in 4k, I managed to do it once, but I couldn't save it, the site crashed, and I only got a glimpse of the result.
- AI rendering is far from ideal in terms of preserving the composition of the original image. Every second generation creates a completely different view related to input, but maybe this is a problem with Google AI Studio, which confuses the context with previous requests. We'll see tomorrow in other services.
- The camera change and image editing works better than in the previous version.
- The price has increased fourfold, but maybe the stable version (not the preview) will work four times more reliably, meaning fewer generations will be needed to achieve the result. There are reasons to believe this.
#gemini
- The level of details is like nothing we've seen before; for some shots it doesn't look like AI at all. After the initial generation, the model can upscale the image without losing details to 2k or 4k, depending on the settings.
- It is not very stable yet; in fact, you can only try it through the API, and not all generations are successful there. I was unable to generate an image in 4k, I managed to do it once, but I couldn't save it, the site crashed, and I only got a glimpse of the result.
- AI rendering is far from ideal in terms of preserving the composition of the original image. Every second generation creates a completely different view related to input, but maybe this is a problem with Google AI Studio, which confuses the context with previous requests. We'll see tomorrow in other services.
- The camera change and image editing works better than in the previous version.
- The price has increased fourfold, but maybe the stable version (not the preview) will work four times more reliably, meaning fewer generations will be needed to achieve the result. There are reasons to believe this.
#gemini
❤4🔥2
878afaed-8312-4ada-befb-1394d847af0d.jpg
3.2 MB
Here is an uncompressed 2k image so you can see the details.
🔥2⚡1
new Gemini has been added to #comfyui (currently available in cloud and portable versions, for desktop versions the update seems to be still on the way), and two templates have been included, one specifically for architects to create drawings from photos or renders. It should be noted that the new model, also specifically with this template, works much better in my tests with a resolution of 1k (if you set it to 2k+, you get a different building in the drawings). It seems that the ideal option right now is to first make a small image, and then ask gemini to keep everything as it is, but set 2k or 4k in the settings.
👍4⚡1👏1
Example: input image and drawings for it. I'm showing the first three random generations; I didn't select them. The prompt is taken directly from the template as well; if you add your own details to it, the result will definitely be better.
❤2⚡1🔥1
#rhino 9 is currently in beta (if you have a Rhino 8 license, you can download it), and so far, most of the new features relate to drawings and annotations: custom lines, hatches, new functions for text, dimensions, and callouts, ceiling plans... you can see the full list here. It seems that the goal of these changes is to attract more people who are still working in AutoCAD.
🔥8👍2
Ubisoft is often hated in gaming communities, but I've respected them since childhood for Prince of Persia (The Sands of Time) and Splinter Cell (the very first one; the next ones came out when I was in university, and I didn't have time for them, but I feel like they weren't as good). And now they've released something for #comfyui that looks absolutely gorgeous based on the description and reviews — a PBR material generator. I haven't tried it myself yet, as today's update of Comfy itself is acting weird with the installation of custom nodes, but I think they'll fix it soon. In general, if you do classic rendering and didn't need all this AI before, or you just don't like it, ComfyUI is now your tool anyway, as there seems to be nothing better in terms of creating/searching for materials.
Here are the details:
https://blog.comfy.org/p/ubisoft-open-sources-the-chord-model
Here are the details:
https://blog.comfy.org/p/ubisoft-open-sources-the-chord-model
👍4
Microsoft has released an impressive model for 3D generation, here is a description and examples, and you can test it on huggingface
Media is too big
VIEW IN TELEGRAM
I made an app using vibe coding for planning a conceptual project, like assigning tasks and tracking them, calculating the number of people in the team and time for each task. I think it turned out pretty cool. Check out the video to see what it looks like, and I'll publish the app itself in a bit, once I finish it.
🔥9👍1
I've been busy not only with my work in the office, but also with my own project (I'll show you later, it's probably interesting), so my sleep has been reduced to 3-4 hours a day, and I haven't had any time left for the channel. But now I have an important reason to write something — a reminder for you that this weekend there will be a workshop that I am leading on the DesignMorphine platform (two days, free, #rhino + #blender + #ai). More than 500 people have already signed up (as usual, far fewer will actually attend, but it's still nice), so maybe you'll want to join too. All the details are at the link:
https://designmorphine.com/education/emergent-futures-v1-0
https://designmorphine.com/education/emergent-futures-v1-0
DesignMorphine - Emergent Futures V1.0
Emergent Futures V1.0
Generative World Building
🔥8❤2👍2🙏1🎄1
A long interesting article about upscale, including tasks related to video, has been published on the #comfyui blog.
https://blog.comfy.org/p/upscaling-in-comfyui
https://blog.comfy.org/p/upscaling-in-comfyui
blog.comfy.org
The Complete AI Upscaling Handbook: All in ComfyUI
Benchmarks, 10 use cases, and 20 workflows across every modern AI upscaling model.
👍4👎1
Media is too big
VIEW IN TELEGRAM
Ira Sorokina and I worked on a project during the Christmas holidays, and it takes a minute and a half just to quickly flip through it. It's essentially an entire city. I'll show the details later.
AI was used, of course, but so were many other tools. The project is real, it has a client, and maybe someday it will be built in whole or in part. #myworkdiary
AI was used, of course, but so were many other tools. The project is real, it has a client, and maybe someday it will be built in whole or in part. #myworkdiary
🔥6❤4👍1🤬1