[ARCHIVED THREAD] - FramePack - Generate AI videos from images using your computer's GPU (Now ComfyUI Megathread!) (Page 8 of 17)
 Right of Way  Right of Way rev.A |
|
Can you get him to shoot the bear? |
|
Quoted: I asked ChatGPT to "clean up" the image, then WAN VACE to have it "drive down the street" https://www.ar15.com/media/mediaFiles/563643/ChatGPT_Image_Jun_7__2025__10_34_34_PM-3558557.png https://www.ar15.com/media/mediaFiles/563643/ComfyUI_00001_-ezgif_com-optimize-3558556.gif |
Animating Rhodesia bush war pics River Crossing Probably would have looked a lot better had I upscaled and sharpened the pic first.... |
|
Quoted: Animating Rhodesia bush war pics  https://www.youtube.com/watch?v=qah7yU5GPtI https://www.youtube.com/watch?v=3IbrJS33xEAProbably would have looked a lot better had I upscaled and sharpened the pic first.... https://www.youtube.com/watch?v=RIXgLMbfmXw https://www.youtube.com/watch?v=H9Lnlyi7bSgGreat stuff as usual! I think they could use some film grain to add authenticity. |
|
Actually did better than I thought given the quality of the image |
|
Quoted: Could someone please animate this pic of my wife’s parents. Both long gone. It’ll make my wife cry with joy. Have them smiling, laughing, and then a sweet loving kiss. https://i.postimg.cc/bryxXJNx/IMG-5177.jpg Bump for help in the above pic. TY. |
|
I'll make an attempt to make it look less saturated. Already using a very low number of steps as-is.  20% fewer steps  photobombed So I was just reading that Causvid is causing the oversaturation... |
Gonna try some Kill Bill-themed stuff The new UTG magnetic wml mount ETA> had no idea that youtube had a daily upload limit  Got a 24 hour time-out, lul. A shame, because I just realized that I have been using nerfed quality settings. |
|
Damn, that's awesome. Reading this thread is fascinating, first I've paid attention to how far all of this has come. |
I made a (terrible) little video of the JGPs I used for the following videos, then merged them all. I spent the last hour trying to figure out how to add audio. My late wife - JPGs to mp4 with AI |
Something a little different. Now using the "full" fp32 model. Better quality, but at a greatly increased cost in generating time. artsy fartsy Don't waste your time trying to psychoanalyze. The prompt was pretty much picked at random. Note to self- turn off face restore mode before generating pic with hundreds of skulls, lul. took forever to complete..... Attempt#2. Much better, I think.  Wan2.1 14B T2V FusionX Make sure to view at 1440 |
|
At the very least, only install from well-known repositories on Github or maybe Huggingface. While I've never tried it, you could probably run it in sandbox mode (through an anti-virus program such as Avast). |
|
Quoted: I made a (terrible) little video of the JGPs I used for the following videos, then merged them all. I spent the last hour trying to figure out how to add audio. https://www.youtube.com/watch?v=aD8IbDLiG7wI hit the like button! I'm glad you could relive your memories through this tool. Hey, your PC is strong enough to game now. Red Dead Redemption 2 seems like your style. Quoted: How secure is this? Installing random AI on PC kinda gives me pause... Like GySgt_D alluded to, if it is on GitHub and Hugging Face and has been there a while with enough downloads, it's probably fine. ComfyUI is perfectly safe and there is really no risk in downloading the latest model and dragging it to its respective folder. |
|
Quoted: Bump because this thread is cool and my PC is to garbage to really participate  Not yet at least! Quoted: At the very least, only install from well-known repositories on Github or maybe Huggingface. While I've never tried it, you could probably run it in sandbox mode (through an anti-virus program such as Avast). https://www.youtube.com/watch?v=jr75N8KrVpgThe skull in the sky looks sick! Lately, I've been playing around with Hidream i1. |
|
Quoted: I hit the like button! I'm glad you could relive your memories through this tool. Quoted: Quoted: I made a (terrible) little video of the JGPs I used for the following videos, then merged them all. I spent the last hour trying to figure out how to add audio. https://www.youtube.com/watch?v=aD8IbDLiG7wI hit the like button! I'm glad you could relive your memories through this tool. Thank you for that, and for helping to make it possible. 
|
|
"graphic novel scratched charcoal sketch of, swirling ancestral ghosts, spirits of ancestors, subtle distant vapor faces of population amidst diffuse spectral storm clouds, scenery millions of skulls in path of oncoming military_tank, driving over foreground mountain of skulls, relentless overpowering emotion with red paint thrown on canvas, of charcoal dust scratched in ash, scarred, cursed promise" |
|
Attached File All five vids contained swastikas, but I'm thinking that the one that got striked may have included one in the thumbnail. "Is there a specific rule you violated?" "Hate" speech 
|
|
Quoted: https://www.ar15.com/media/mediaFiles/3906/Screenshot_2025-06-25_at_05-11-53_Channe-3572392.JPG Daily reminder of who actually rules over us Is there a specific rule you violated? I know you are supposed to add a disclaimer stating the video contains AI generated content. There are channels that have brain rot AI generated videos for kids who don’t know the difference. |
|
I found a guy on Hugging Face that makes .bat file scripts to install Triton and Sage Attention a automatic thing. He even includes NVIDIA Apex with it so you don't have to worry about installing CUDA and cuDNN if you don't want. Make sure you know what models you want before hand as the installers lets you choose what ones you want. I chose all the WAN ones and it was quite the download wait.... ComfyUI-Auto_installer |
|
Quoted: Is there a specific rule you violated? I know you are supposed to add a disclaimer stating the video contains AI generated content. There are channels that have brain rot AI generated videos for kids who don’t know the difference. Quoted: Quoted: https://www.ar15.com/media/mediaFiles/3906/Screenshot_2025-06-25_at_05-11-53_Channe-3572392.JPG Daily reminder of who actually rules over us Is there a specific rule you violated? I know you are supposed to add a disclaimer stating the video contains AI generated content. There are channels that have brain rot AI generated videos for kids who don’t know the difference. The only conversation should be about what was uploaded. YouTube doesn't have an issue (at all) with AI generated content. |
|
I guess battlefield art involving Soviets is A-OK, but not Germans. Also animated some WWII war-time posters "loose lips sink ships", etc. Scared to upload those, lul.  non-offensive video of a street in Liberalville This is using the fp16 version of the model. Fp32 looks better (although somewhat oversaturated), but takes 3x as long to generate. Another fp16 example (374 seconds to generate)-  fp16 example The same, but in fp32 (903 seconds to generate)- I don't think I see any benefit to the latter, but I probably shouldn't have done this comparison in such low resolution... ETA> Had much better results by simply increasing resolution from 720x to something like 1040. Go figure. |
|
Quoted: I have yet to try VACE. Don't even know what it is supposed to do, actually. Attempted to use a VACE workflow a couple weeks ago, but the documentation for it was the usual awful to non-existent VACE CONTROLNET Workflow Step-by-Step Guide Series: WAN - VACE CONTROLNET |
|
Quoted: I found a guy on Hugging Face that makes .bat file scripts to install Triton and Sage Attention a automatic thing. He even includes NVIDIA Apex with it so you don't have to worry about installing CUDA and cuDNN if you don't want. Make sure you know what models you want before hand as the installers lets you choose what ones you want. I chose all the WAN ones and it was quite the download wait.... ComfyUI-Auto_installer I deleted my ComfyUI installation and tried the auto installer. It didn’t work. Besides, that appears to be the portable version which has some drawbacks. I really wish I could get Sage Attention and Triton to work. |
|
Quoted: I deleted my ComfyUI installation and tried the auto installer. It didn't work. Besides, that appears to be the portable version which has some drawbacks. I really wish I could get Sage Attention and Triton to work. What problem did you have with the auto installer. If you pick a wrong combination of model, vae and clip it will output as blank. He has more workflows on his Huggingface site you can try if the included ones don't work. |
|
Quoted: Yeah I"ve done all the install ways, miniconda, python pinokio, portable and the new installer. Miniconda was my fav to work with but getting SA and Tri to work was a pain in the ass so I gave in.... What problem did you have with the auto installer. If you pick a wrong combination of model, vae and clip it will output as blank. He has more workflows on his Huggingface site you can try if the included ones don't work. Quoted: Quoted: I deleted my ComfyUI installation and tried the auto installer. It didn't work. Besides, that appears to be the portable version which has some drawbacks. I really wish I could get Sage Attention and Triton to work. What problem did you have with the auto installer. If you pick a wrong combination of model, vae and clip it will output as blank. He has more workflows on his Huggingface site you can try if the included ones don't work. The installer was crashing but I figured out I had an unstable PBO overclock. When I used to have the portable version, it wouldn't let me install custom nodes. Anyway, I got the regular version installed again without Teacache or Sage Attention. Flux Face Swap is fun    
|
|
Quoted: I guess battlefield art involving Soviets is A-OK, but not Germans. Also animated some WWII war-time posters "loose lips sink ships", etc. Scared to upload those, lul. https://www.youtube.com/watch?v=dhCIoCMfRR8This is using the fp16 version of the model. Fp32 looks better (although somewhat oversaturated), but takes 3x as long to generate. Another fp16 example (374 seconds to generate)- https://www.youtube.com/watch?v=_LLylmbjGscThe same, but in fp32 (903 seconds to generate)- https://www.youtube.com/watch?v=E556Kzu4YD0I don't think I see any benefit to the latter, but I probably shouldn't have done this comparison in such low resolution... What were you trying with the Russians that worried but not for German version? I suppose Russia was on opposing sides and not happy about it due to the meat grinder they had going on the Eastern Front up north. Or is it a "westernized" Germany that needs to be appeased while there's no pressure regarding Russia/USSR. There are pretty cool to give a glimpse of monochrome but in color. Reminded of the song by Carbon Trace "War was in Color" which is sort of a one hit wonder for them. |
 Red, White, and Blue This was originally created earlier this year using Hunyuan, iirc. Now animated using Wan. |
|
Nice! I’m impressed with how realistic FramePack can be when you use a prompt that works within the model’s limitations. Look at how the fire reacts when she moves the embers. |
|
New P1 version of FramePack is coming out. Here is the preview page: https://lllyasviel.github.io/frame_pack_gitpage/p1/ |
|
Quoted: Animated Lootie! What a time to be alive. somebody should @jdissel not sure if that's the right username or not. The OG who found real life lootie. What an epic thread that was. We argued for like a dozen pages describing his ear lobes and whether it was really him or not. The good ol days... One of the classic Arfcom threads. |
|
I finally got Triton, SageAttention2 and Multi GPU nodes working in ComfyUI. Hidream i1 Text to image of "S2000 driving through the highlands of Scotland" followed by WAN 2.1 FusionX image to video with the same prompt. The 5-second video took 95 seconds to generate. 
|