Comfyui support ?
any ETA ?
It's based on LTX, so it shouldn't need many fixes. ComfyUI is already trying to run it, but I’m not sure exactly what needs to be changed yet.
Gguf version??
love smthem but he never quants to low sizes. i cant run a 22gb model on 8gb of vram and he wont be up until were asleep in china now lol
love smthem but he never quants to low sizes. i cant run a 22gb model on 8gb of vram and he wont be up until were asleep in china now lol
Are you even tried after DynamicVRAM introduction? Its Unified memory like on Macs, Vram no matter anymore, system Ram matter. If you have enough system Ram try to launch the max size LTX model, it will run and completes (better to have xformers for speed, it really helps).
love smthem but he never quants to low sizes. i cant run a 22gb model on 8gb of vram and he wont be up until were asleep in china now lol
Are you even tried after DynamicVRAM introduction? Its Unified memory like on Macs, Vram no matter anymore, system Ram matter. If you have enough system Ram try to launch the max size LTX model, it will run and completes (better to have xformers for speed, it really helps).
Honestly, smthemex is a standalone loader; it's not used like ComfyUI. If use Comfy's loader, there won't be any problems.
@Kimmypox @krustik @PushkarHF1 @realrebelai @ryg81
Hi all, ComfyUI support for JoyAI-Echo is now available.
You can use our custom node here: https://github.com/zhuang2002/ComfyUI_JoyAI_Echo
This implementation includes complete DMD denoising settings, audio-video memory mechanism and correct parameters, which can reproduce the standard video quality perfectly.
@Kimmypox @krustik @PushkarHF1 @realrebelai @ryg81
Hi all, ComfyUI support for JoyAI-Echo is now available.
You can use our custom node here: https://github.com/zhuang2002/ComfyUI_JoyAI_EchoThis implementation includes complete DMD denoising settings, audio-video memory mechanism and correct parameters, which can reproduce the standard video quality perfectly.
i have ggufs, working on pulling the connectorr file from the model so we can run it with a gemma encoder and seperate the workflow. node implementation is currently coding by claude to add into your repo as well for utilization of my quants.
https://huggingface.co/realrebelai/JoyAI-Echo_GGUF/tree/main
@Kimmypox @krustik @PushkarHF1 @realrebelai @ryg81
Hi all, ComfyUI support for JoyAI-Echo is now available.
You can use our custom node here: https://github.com/zhuang2002/ComfyUI_JoyAI_EchoThis implementation includes complete DMD denoising settings, audio-video memory mechanism and correct parameters, which can reproduce the standard video quality perfectly.
i have ggufs, working on pulling the connectorr file from the model so we can run it with a gemma encoder and seperate the workflow. node implementation is currently coding by claude to add into your repo as well for utilization of my quants.
https://huggingface.co/realrebelai/JoyAI-Echo_GGUF/tree/main
Thanks for your work on this! Feel free to open a PR when it's ready — we'd be happy to review it.
My friends, I really don't get why the Comfy team seems so resistant to these ultra-long, consistency-maintaining models. From the "StoryMem" project back in the Wan2.2 days to "JoyAI" now, I’ve opened support requests on their GitHub, but it feels like they haven't even formally reviewed them. It honestly makes you wonder if there’s some conflict of interest going on.
Regardless, I’d still really appreciate it if you guys could show some support!
I’ve opened support requests on their GitHub, but it feels like they haven't even formally reviewed them.
Your issue is one of many. As far as I know Comfy is always interested if the developers come to them before their release (but may be there's is also the thought what the LTX developers from Lightricks think about it, who knows? I don't want to speculate).
It honestly makes you wonder if there’s some conflict of interest going on.
Currently they've integrated Ideogram4 (before that e.g. Ernie Image, Microsoft Lens and HiDream 01 in May), they constantly work on such things. I'm just an user but the expectations some people have about an open-source software is highly unrealistic. I don't believe it's about being resistant to "ultra-long, consistency-maintaining models". Every week there are new models, there are checks what fits and what not.
@tuolaku , I'm just testing smthemex's ComfyUI custom-node for JoyAI-Echo ( https://github.com/smthemex/ComfyUI_JoyAI_Echo ), that works well (the Q8 22 GB model, on my little 8 GB VRAM, 64GB system RAM), this could be also an option for you. It comes with a workflow, I had to translate the Chinese notes, but the first run already had no errors. Now I look into jdopensource's infos to see how prompts are different (single prompt JSON). I'll later test @realrebelai 's JoyAI-Echo-DiT-Q4_K_M.gguf model.
I’ve opened support requests on their GitHub, but it feels like they haven't even formally reviewed them.
Your issue is one of many. As far as I know Comfy is always interested if the developers come to them before their release (but may be there's is also the thought what the LTX developers from Lightricks think about it, who knows? I don't want to speculate).
It honestly makes you wonder if there’s some conflict of interest going on.
Currently they've integrated Ideogram4 (before that e.g. Ernie Image, Microsoft Lens and HiDream 01 in May), they constantly work on such things. I'm just an user but the expectations some people have about an open-source software is highly unrealistic. I don't believe it's about being resistant to "ultra-long, consistency-maintaining models". Every week there are new models, there are checks what fits and what not.
@tuolaku , I'm just testing smthemex's ComfyUI custom-node for JoyAI-Echo ( https://github.com/smthemex/ComfyUI_JoyAI_Echo ), that works well (the Q8 22 GB model, on my little 8 GB VRAM, 64GB system RAM), this could be also an option for you. It comes with a workflow, I had to translate the Chinese notes, but the first run already had no errors. Now I look into jdopensource's infos to see how prompts are different (single prompt JSON). I'll later test @realrebelai 's JoyAI-Echo-DiT-Q4_K_M.gguf model.
smthemex's ComfyUI nodes have been updated and are ready to use. I'm currently testing them myself, and I can confirm that 6GB of VRAM was enough in the previous version. This is super helpful for users with limited video memory.
@realrebelai 's JoyAI-Echo-DiT-Q4_K_M.gguf model (who seems to work on a node) doesn't work with smthemex's ComfyUI node. I don't test the Comfy node https://github.com/zhuang2002/ComfyUI_JoyAI_Echo because it expects far more RAM than I have. Of course it would be easier for the VRAM poor among us if Comfy could find a solution to use JoyAI-Echo simplified. If every ComfyUI custom-node comes with their own specific models (and folder paths) it's rather unappealing for my harddrive.
Third-party ComfyUI nodes can't fully take advantage of ComfyUI's VRAM management system, which is one of the main reasons I've been hoping for official support from the ComfyUI team. After all, great models deserve to be accessible to more people.
My implementation for ggufs is almost done. Just running big fixes at this point on breaks at work, i have a full time im currently stuck at with my laptop and chrome desktop so bare with me haha
My implementation for ggufs is almost done. Just running big fixes at this point on breaks at work, i have a full time im currently stuck at with my laptop and chrome desktop so bare with me haha
you have been killing it lately bro, thanks