Prompt following leaves something to be desired.

Alternative Euler-A, Steps 10, CFG 1.0, 1024x1024, model z_image_turbo_bf16, seed 1, prompt "A small angry squirrel pursues a big frightened fleeing cat."

Z-image likes detailed prompts on what to make exactly, the encoder has difficulty to interpret vague terms like 'pursues'. Ask AI like CoPilot to write a "script" for you.

Example prompt:
Scene: A dramatic autumn garden path, scattered with orange and yellow leaves.
Action: A small, furious squirrel is mid-air, leaping aggressively with claws extended and tail bristling. Its eyes are locked on a much larger, terrified tabby cat fleeing ahead. The cat glances back in panic, ears flattened, mouth open, legs stretched in full sprint.
Composition: The squirrel is clearly behind the cat, closing the gap. The cat is ahead, off-balance, with motion blur and flying leaves emphasizing speed and fear.
Emotion: The squirrel radiates rage and determination. The cat shows pure panic and desperation.
Style: Cinematic lighting, low-angle perspective, shallow depth of field. Emphasize dynamic motion, emotional contrast, and predator-prey tension.

Result:

--

--

The "wildcard" that CoPilot suggests are options for the generator. "Picture of a cat" generates a cat, "Picture of a {cat|dog|elephant|squirrel}" will randomly pick an animal. CoPilot can generate a base prompt with multiple wildcards. You then can run some batches and see what images fit your need.

Example prompt:
A {squirrel|mouse|duck|frog|hedgehog} {furious|determined|mischievous|gleeful} {chases|lunges at|sprints after|bounds toward} a {cat|dog|goose|rabbit|fox} {terrified|panicked|startled|confused} {flees|scrambles away|bolts|runs in panic} through a {snowy|rainy|foggy|sunny|windy|overcast} {garden path|forest trail|city alley|grassy field|suburban street}.
The chaser is behind, mid-pounce with {claws extended|tail bristling|mid-leap|eyes locked on target|mouth open}, while the runner is ahead, {ears flattened|glancing back|legs stretched|tail raised|fur standing on end}. Flying debris adds urgency: {snow spray|rain splashes|falling leaves|dust and grit}.
Rendered as a {photo|painting|clay figures|digital illustration|cartoon|oil on canvas}, with {dynamic motion|predator-prey tension|emotional contrast|cinematic framing|playful chaos} and {cinematic lighting|soft natural light|dramatic shadows|warm glow|diffused daylight|golden hour tones}. Shot from a {low angle|eye level|tracking side view} with {motion blur|shallow depth of field|panning streaks}.

Result:

Yeah I hit that error as well. I'm using a custom comfyui server, so I updated ComfyUI and comfyui_tooling_nodes on my custom server . I was able to render after that.

I still haven't been able to solve it.

In case you haven't tried yet:

kijai/ComfyUI-WanVideoWrapper#793

Open Preferences -> Styles, select Z-Image Turbo from the dropdown. If it says "checkpoint not installed", update ComfyUI and comfyui-tooling-nodes. Restart ComfyUI and Krita.

thank you,I successfully ran KRITA Z IMAGE using your method.

релиз ещё не вышел.

если вы скачаете последнюю версию comfyui и последнюю версию плагина krita из git, то сможете их использовать

Please tell me what exactly needs to be updated, comfyUI or manager to the latest version? What version exactly do you have at the moment, that Krita AI has the ability to support controlnet for z-image?

How did you install it, in the first place?

I downloaded KritaAI from GitHub and installed it last year. I hadn't used Comphy's internal nodes until last week, but for optimal performance with ZIT, I need a newer version of Comphy and, preferably, Manager to update new nodes.
I don't know how to update this "built-in" Krita Comphy now.

I've never used the managed server myself, but from what I can tell, the plugin just asks for a folder to install ComfyUI into. You should be able to cd into that folder and update everything from there.

yes, z-image support was added to version v0.3.75 of comfyui with further improvements at v0.3.76 , check the comfyui release notes here https://github.com/comfyanonymous/ComfyUI/releases

Hey @Acly, You might want to update the post since they just dropped an new and recommended 8step variant of the version 2.1 of both the tiled and union controlnets with claimed better quality and clarity at distilled speed.

Yea I was just testing it.

Compared to the non-distilled version it no longer looks blury and washed out, but IMO it tends to overshoot a bit into the other direction. It looks a bit fried like what you get when you run a 8-step model with too many steps.

Oh, that's a bummer , so can that be remedied by reducing the controlnet strength? , or do we still need a way to set a step range to prevent that from happening?

It seems that my issue was that I had put the file into the ai_diffusion/server/models/text_encoders folder and NOT the ai_diffusion/server/ComfyUI/models/text_encoders folder. It actually drives me nuts that both of these exist and I never know which is the correct one.

ai_diffusion/server/models/text_encoders is correct and works for me

don't put it into ComfyUI folder, it also works, but will prevent updates (which replace ComfyUI with new version)

Good to know, but it did not work for me until I put the models there. I dunno why, but it is working now and would not before I moved them.

I am having the exact same issue. Just to mess around I chose to upload the Fun-Controlnet safetensors file as lora for Z-Image and set it to around 20% it actually changes the inpaint outcome. Still I was not able to implement control layers but I was just feeling adventurous 🤣

It looks to me that there's something wrong with the backend. Did you look at the client.log or server.log? Is this local environment or remote ComfyUI server?

I looked at it, but I don't understand programming very well...I don't know what exactly to look for there. Moreover, Krita itself pre-installs and controls everything necessary. As for ComfyUI, I used a local Krita server, but now I have installed a portable ComfyUI separately to work through a configurable server in order to expand the possibilities of interaction. I haven't checked yet whether the full Z-Image-Turbo model works via portable ComfyUI. But I found a temporary solution for now by downloading the GGFU version of Z-Image-Turbo.

Thanks for the update. On Mac M4, I'm trying to use the lite variant but I'm hitting this error:

size mismatch for control_all_x_embedder.2-1.weight: copying a param with shape torch.Size([3840, 132]) from checkpoint, the shape in current model is torch.Size([3840, 64]).

FYI - I've updated the plugin, tooling and inpaint nodes and both 2601 control nets are both in model_patches What am i missing? TIA

needs comfyui update too

I'm one of those ComfyUI Desktop on Mac weirdos. I think Desktop lags behind ComfyUI versions, so I'll have to wait. But non-lite versions work great.