Made a cli tool for running z-image-turbo, turns out it's great for making political cartoons because you can provide a detailed scene description using JSON
it's also great for making political cartoons, and you can generate the initial json description using something like deepseek, then tweak it after to get the result you want
haui - 2w
Can we make a community where comrades who are creative ask comrades who are into ai to help with making propaganda? We could also make it a propaganda suggestion place.
I recently had this image in front of my eyes where workers, women, children, etc. are shredded by a giant paper shredder called capitalism and it drops coins and dollar bills.
If someone wants to run with this, feel free.
3
☆ Yσɠƚԋσʂ ☆ - 2w
here's a version of the idea I managed to get with the script
6
haui - 2w
DAMN! That rocks! Thank you!
1
☆ Yσɠƚԋσʂ ☆ - 2w
oh and just updated it, got an even better version
1
haui - 2w
Looks great. Thanks for sharing.
2
CriticalResist8 - 2w
We could extend this community to asking questions about using local AI tools too. How much Vram do you have on your machine? That will determine if you can start generating your own images! (The rest comes with practice)
3
haui - 2w
I have 8 gigs of vram.
1
CriticalResist8 - 2w
With 8GB you could run the Q8 gguf of z-image turbo (same model yog used in this post): https://huggingface.co/jayn7/Z-Image-Turbo-GGUF/tree/main! So it's definitely doable on your machine with ComfyUI. It might be a bit slow or make your computer freeze at times so if that happens you can also try the Q6, the difference in quality is not super different.
To install ComfyUI I will probably post a guide but, in the meantime while I set that up, you can send deepseek web the link to their github: https://github.com/comfyanonymous/ComfyUI and ask it to write a guide to install it on your machine.
4
haui - 2w
Thanks for the suggestion. I'll save this and come back to it. I'm really curious what this might achieve in combination with revolutionary ideology.
1
CriticalResist8 - 2w
Okay, took a while to get it to work since I had to install comfyUI on my new OS, but I got there. Quick guide:
-Get CUDA/Metal/AMD thing first if you don't have it. Easy enough with commands iirc.
Put the downloaded models in the correct folders in ComfyUI/models/[folder]. So for example on huggingface repo there's a vae folder and an ae.safetensors file in it, this means you will put that ae.safetensors file in ComfyUI/models/vae/ae.safetensors.
I renamed my ae.safetensors to zimage-vae.safetensors. To download on huggingface click the small arrow pointing down next to the model's name in the list.
Open comfyUI, Ctrl+O, select the workflow json. Note: if you renamed the vae or any other file, first edit all instances in the json. I don't know why but it wouldn't work otherwise.
Very simple workflow but it works, got a 2048x2048 in about 1 minute if I'm reading the output correctly which is both very impressive and very scary (edit: right click the big node where you put the prompt in and do "uncompress node" or something like that then rearrange the new nodes, otherwise it will use hardcoded values and you won't be able to get any other image than the default)
This huggingface repo needs about 15-16GB of Vram to run.
Will upload a sample image in a few minutes... just debugging some other stuff first.
Edit: here you go, minified beforehand with curtail's lossy compression algorithm so it is a tiny bit different from what z-image came up with. Didn't change anything from what came with the simple workflow I linked. You could use loras and upscalers but comfyui is complex enough for me as it is lol.
yogthos in crushagent
Made a cli tool for running z-image-turbo, turns out it's great for making political cartoons because you can provide a detailed scene description using JSON
https://git.sr.ht/~yogthos/z-image-turbo-cli
You can see an example of a scene here https://git.sr.ht/~yogthos/z-image-turbo-cli/tree/main/item/input.json
it's also great for making political cartoons, and you can generate the initial json description using something like deepseek, then tweak it after to get the result you want
Can we make a community where comrades who are creative ask comrades who are into ai to help with making propaganda? We could also make it a propaganda suggestion place.
I recently had this image in front of my eyes where workers, women, children, etc. are shredded by a giant paper shredder called capitalism and it drops coins and dollar bills.
If someone wants to run with this, feel free.
here's a version of the idea I managed to get with the script
DAMN! That rocks! Thank you!
oh and just updated it, got an even better version
Looks great. Thanks for sharing.
We could extend this community to asking questions about using local AI tools too. How much Vram do you have on your machine? That will determine if you can start generating your own images! (The rest comes with practice)
I have 8 gigs of vram.
With 8GB you could run the Q8 gguf of z-image turbo (same model yog used in this post): https://huggingface.co/jayn7/Z-Image-Turbo-GGUF/tree/main! So it's definitely doable on your machine with ComfyUI. It might be a bit slow or make your computer freeze at times so if that happens you can also try the Q6, the difference in quality is not super different.
To install ComfyUI I will probably post a guide but, in the meantime while I set that up, you can send deepseek web the link to their github: https://github.com/comfyanonymous/ComfyUI and ask it to write a guide to install it on your machine.
Thanks for the suggestion. I'll save this and come back to it. I'm really curious what this might achieve in combination with revolutionary ideology.
Okay, took a while to get it to work since I had to install comfyUI on my new OS, but I got there. Quick guide:
-Get CUDA/Metal/AMD thing first if you don't have it. Easy enough with commands iirc.
This huggingface repo needs about 15-16GB of Vram to run.
Will upload a sample image in a few minutes... just debugging some other stuff first.
Edit: here you go, minified beforehand with curtail's lossy compression algorithm so it is a tiny bit different from what z-image came up with. Didn't change anything from what came with the simple workflow I linked. You could use loras and upscalers but comfyui is complex enough for me as it is lol.
better sampler (dpmpp_sde and ddim_uniform)