Keeper of Ash

@DeepPromptCine I pretty much have to mute the sound before I start to watch my generations with grok hehe, it's a big problem! :u

@keeper_of_ash Yeah i tend to agree with the consensus that the audio has that Grok "signature" sound. They need to get an audio engineer onboard or partner with ElevenLabs.

There's a new mystery model on Artificial Analysis called Purple 👀 artificialanalysis.ai/video/arena Only shows up under Text-to-Video + With Audio. Spent an hour hunting and managed to pull 4 (got a bit unlucky). It actually seems to be a very strong model, not perfect but really good with slower videos, and VERY CRISP image quality, The weak point is the audio, and there's a very annoying noise in the background almost all the time. Visually good with details and physics. What do you guys think? and who may be behind the model? Check out the videos here. @grok what model do you think it is? :D don't be shy, you have to make a good guess! And then also tell me if you think the model is better than your new Grok Imagine 1.5 model. I bet you will admit it's superior to yours 😏 still behind Seedance 2.0 and now this one as well?😅 #newaimodel #videomodel #TextToVideo #videoarena #ArtificialAnalysis

@pwkwjwt it lures them in 🐾😽

猫が入って来ないようにスパイク付きのフェンスにしました

@BrentLynch I actually like the model because it's different, and sometime very good quality, may be a good addon to explore, just got an idea. Gonna try to create pictures with reve 2 and then later modify them, improve them with GPT Image 2. gonna try it out now :P

AI DROPS ARE BACK! First IDEOGRAM 4 Now REVE 2. My initial impressions? It's improved from the last Reve release? Did they just kill (INSERT TOP AI MODEL HERE) No, but I don't think Reve is meant for sword fighting and gunslinging AI ladies use cases. :)

@STFUnion You never know with grok, or musk :P. but I strongly agree! very fun with new models! I'm tired of Seedance 2.0 dominance at the moment. looking forwar to a price war :D

@keeper_of_ash That's a very good point, but also could be the reason they didn't release T2V because of those problems. Either way, I'm very excited for another solid model. It seems video is getting close again!

@STFUnion thats true, I was thinking about it a little bit myself but I actually find this sound even worse than what grok is now, even with the 1.0 version. would be a bit strange to have worse sounds with text to video compared to picture to video.

@keeper_of_ash I mean it wouldn't be that surprising if it was the text to image version of the new Grok Model. The current new Grok model is only image to video. I could very well be wrong just seems odd that this new model is specifically text to video.

@BrentLynch I catched a few more purple ones :) and yes, the sound quality seem to be bad, on pretty much everyone.

👀WHAT SECRET AI VIDEO MODEL IS THIS? IS IT HAILUO 3? IS IT SEEDANCE 2.1? IS IT KLING? What do you think? Goes by the name of PURPLE and just landed on Artificial Analysis. It's GOOD! Spent about an hour pulling some additional samples and will post some more samples in thread below. A vintage car's curved chrome bumper reflecting a distorted fisheye view of a 1950s gas station. As someone polishes the bumper with a cloth, the reflection clarifies temporarily before fingerprints and cloth streaks create new distortions. The pump attendant's reflection stretches and compresses as he moves around the vehicle. Audio: Cloth buffing chrome, distant traffic, vintage radio music. 🔽

@HauntedAI I never expected a reddit moderator to look anything different than this, spot on. and his own fantasyworld spot on as well! He is the best, and You know it :)

Being a Reddit moderator is the most noble profession.

@BrentLynch Love to see new models pop up! hopefully the era of Seedance 2.0 dominance is soon over, does anyone know if the new model is good at action as well or just slower cinematic shots?

Been testing the two newest image models out right now: Reve 2.0 and Microsofts mai-image 2.5 Compared them against the other top picture models out there - GPT Image 2 and Nano Banana 2. Each model got 3 generations, with exactly the same prompt. I kept the best picture from each model. Here are the results! #reve2 #maiimage25 #newaimodels #text2image #aiart

@elonmusk for us plebs out there who can't afford the Grok Imagine 1.5 API, and waiting it to be released on grok.com , you can still create pretty good videos now on the 1.0 version as well. I think they have recently updated it :) here is a short one from 1.0

Iliad (Troy) trailer made by Grok Imagine 1.5, which was just released

ok last reply from me here, just want to share my best pictures from this model I generated. Dont expect to get great pictures all the time, some results are great others very bad. and with this model it's even more extreme. for me, gpt image 2.0 feels like you get solid outcome every time, reve 2, you get very mixed instead, all over the place but sometimes super interesting ones. Great model to have on top of other models to get unique experiences. and btw, I have not tried out the "edit image" stuff, that may help with some of the bad generations.

Today, we’re launching Reve 2.0, the best 4K image model in the world. We invented a new way to generate and edit any image using precise layouts. For the first time, it’s possible to create images you can touch.

@oran_ge @reve I tried it now for free, app.reve.com and here is the prices I found, I have not tried the paid version but I can create about 10 pictures, high res (like 4k) after that it it told me to wait 1 day.,so if the info here is right it seem to be pretty good pricing

@reve What’s the price?

@reve Very interesting model, it likes detailed promts, good with negative prompts. The pictures feels different, but very much hit and miss. here are more examples compared to other top models. (used same promts on all models) Reve 2 vs GPT Image 2, Nano Banana 2 and mai image 2.5

@reve I just tested Reve 2.0 against the current top models using the exact same prompt (best out of 3 tries for each model) Compared to GPT Image 2.0, Nano Banana 2, and the brand new model mai-image 2.5 Here are the results.

@anditiarizki I wonder what makes seedance 2.0 so good with the motions, if its bytedance tiktok training data or just some algorytmh secrate sause. I realtid hope some new models can get as good with the motions as seedance 2 soon. Grok imagine 1.5 does not really cut it.

I'm starting to understand why so many people are getting hooked on Seedance, Not because of the AI itself. But because anyone can now create visuals that felt impossible just a few years ago. Anyone else feeling the same way?

@abxxai btw, look at this stuff youtube.com/watch?v=Rjvldl… Great AI short film, I say Holywood is already fucked right now, with the tools we have and it's only gonna be better. I'm so tired of holywoods slop, when true artists can create amazing stuffs soon!

Hollywood is cooked if Seedance 2.0 keeps doing this.😱 One red line = a full cinematic flight. First-person POV. Physics-accurate motion. Fog, lighting, atmosphere. All instant. AI video just entered a new era.

@abxxai does this work with any other models or just seedance 2.0? it's impressive!

@mlech26l 20k GPU rig may sound expensive but it gives free heat, claude can't beat that. gigaslop station ftw

why pay $20/month for Claude if you can run gigaslop_27b_q8 on a $20k GPU rig in your living room

@krea_ai @ArtificialAnlys I recently made comparison pictures of new models and Krea 2 (text-to-image) as well because of the interesting open source release news, posting them here so people can see how this model is doing. I used the same prompt for all models, and only 1 take on each generation.

Krea 2 is now on @ArtificialAnlys #1 image model from an independent research lab and #6 globally on text-to-image leaderboard. open-source cooking and coming soon.