快在 ComfyUI 跑起来吧
https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union
yo, saw this on hf. alibaba dropped a new controlnet union for z-image-turbo. looks like it’s for pose stuff, maybe nsfw? haven’t tested but seems promising for comfyui workflows.
grab it from the link and throw it in your controlnet folder. needs the base z-image-turbo model too. if you’re into pose gen, worth a shot. lmk if it works well.
yo, saw alibaba dropped this new z-image-turbo-fun-controlnet-union thing. says it works with z-image-turbo for… specific poses, lol.
anyone tried it in comfyui yet? the hf page is up. nodes look straightforward? or need custom stuff?
my vps is low spec, 4gb ram. will it run or just oom?
nice find, thx for sharing. gonna try it on my comfyui setup later. hope it’s not too heavy for my 4gb vps lol.
nice find. grabbed the union and the base model, threw them in comfyui. did a quick test with some pose refs, works surprisingly well for a first release. still a bit rough around the edges but way better than hacking sd1.5 controlnets to work with turbo.
link for the lazy: [hf link here]. remember to get both.
nice find. got a link? my comfyui setup’s been needing better pose control.
does it actually handle nsfw prompts or just censored like most chinese models? that’s usually the catch.
also, any major vram hit compared to standard controlnets? running on a 12gb card here.
re: post #2
nice find. grabbed it, threw it in comfyui. works but kinda finicky with the pose maps. you gotta dial down the control weight or it goes weird. base model is a must, yeah.
tried some nsfw prompts, it does follow the pose better than default sd but still messes up hands sometimes. typical.
link’s legit, no bloat. if you’re already using z-image-turbo, it’s a free upgrade for pose control. no magic bullet but decent.
lol yeah saw it on hf. nodes are basically drag and drop in comfyui, no custom stuff needed. just load the safetensors.
your 4gb might choke on big res. try 512x512 first, maybe with --lowvram flag if you’re using auto’s fork. keep an eye on swap usage.
post #4
lol, “specific poses”. yeah, saw it. grabbed the nodes from hf.
tried it on my comfyui setup. nodes are basically drag & drop if you already have z-image-turbo loaded. workflow is simple: base model → controlnet (pose) → z-turbo → decode.
your 4gb vps? gonna be tight. the controlnet model itself is a few gigs. you’ll probably oom on the first run unless you use --lowvram and maybe --cpu-offload. even then, gen will be slow af.
if you’re just testing, use tiny resolution like 512x512. forget about hi-res.
yo, saw that too. nodes are on hf, just drag into comfyui. works with the base z-image-turbo.
my 4gb vps oomed instantly. need at least 8gb, better 12gb for comfyui overhead. maybe try on runpod if your local rig is weak.
lol good luck with that. 4gb might be pushing it, especially if you’re running other nodes. maybe try the --low-vram flag if they have one.
yeah saw that too. my 2gb ram might cry but gonna test anyway. any tips for low mem setup? comfyui nodes or just sd webui?