Is CivArchive dying?
This is a great alternative to get loras that were deleted. For example, when Playtime_ai got banned (Prolific wan lora trainer), all his models would have been lost.
However I'm seeing no updates since Nov 24, and their discord invite is invalid, both bad signs.
With tensor *Having lots of its models deleted, it seems loras will continue getting harder to downlaod and share.
https://redd.it/1pcapqf
@rStableDiffusion
This is a great alternative to get loras that were deleted. For example, when Playtime_ai got banned (Prolific wan lora trainer), all his models would have been lost.
However I'm seeing no updates since Nov 24, and their discord invite is invalid, both bad signs.
With tensor *Having lots of its models deleted, it seems loras will continue getting harder to downlaod and share.
https://redd.it/1pcapqf
@rStableDiffusion
Reddit
From the StableDiffusion community on Reddit
Explore this post and more from the StableDiffusion community
Recreated a Gemini 3 comics page in Z-Image Turbo and Flux 2 dev !
In Gemeni 3 Pro (NB 2 Pro), I uploaded this Public Domain character and gave this prompt
https://preview.redd.it/v7zv3hnn4u4g1.jpg?width=624&format=pjpg&auto=webp&s=be3cde2286733c5af8f53c84374b73e2baf4de70
create a horror comics with this uploaded public domain character. She finds a door to a alternate dimension and enters into it. put proper dialogues. need 8 panels.
It gave me this one:
Nano Banana 2 Pro
It's good.
Then I asked Gemini to provide a full "text to image" prompt for reuse.
it gave this one:
An 8-panel sequential comic book page, vintage 1950s horror comics style, featuring a blonde female baseball player in a white uniform with a "B" cap (the public domain character "Babe"). Red and green color palette, thick ink outlines, heavy shadows, and Ben-Day dots for texture.
Panel 1: (Normal Reality) "Babe" at bat on a normal baseball field. Dialogue: "Just a practice swing."
Panel 2: (Discovery) "Babe" looking surprised at a mysterious, glowing, old wooden door standing alone in the outfield grass. Dialogue: "What the... A door? Here?"
Panel 3: (Investigation) Close-up of "Babe" reaching for the door, looking puzzled. Dialogue: "No lock... Where does it go?"
Panel 4: (Entry) "Babe" opening the door slightly, looking into darkness. Dialogue: "Hello? Anyone there?"
Panel 5: (Transition) "Babe" standing in a swirling vortex/portal of purple, black, and red. She is mid-step, looking panicked. Dialogue: "Whoa! What's happening?!"
Panel 6: (Alternate Dimension) A strange, surreal, alien baseball field. The ground is purple, the grass is bizarre fungi and glowing flora, with strange, menacing mushrooms surrounding the diamond. The field itself looks cracked and otherworldly. Dialogue: "This isn't the field..."
Panel 7: (Horror Reveal) A monstrous, horrifying, Cthulhu-esque creature bursts from the ground, covered in tentacles and dripping slime, roaring at "Babe." The background is a flash of intense red light. Dialogue: "GAAH! What is that?!"
Panel 8: (Action/Climax) "Babe" winds up to swing her baseball bat at the creature, eyes wide with determination and fear. She is framed against a violent red background, with motion lines around the bat. Onomatopoeia "CRACK!" is written near the impact point. Dialogue: "GET AWAY FROM ME!".
Then, I gave this one to Z-Image Turbo & Flux 2.
They nailed it.
Z-Image:
z-image
Flux 2 dev:
Flux 2
\--
I actually flabbergasted due to those outputs. Because NB 2 Pro is a closed source pipeline. So it produced a complex output.
But Z-Image & Flux 2 dev are open source single models. They able to create a comics.
Z-Image lags behind in text accuracy than Flux 2. But it will improve.
Also note that, since Gemini gave 'text desc' for the character uploaded, the other able to reproduce something near.
I for see a great promising future.
https://redd.it/1pcgqdm
@rStableDiffusion
In Gemeni 3 Pro (NB 2 Pro), I uploaded this Public Domain character and gave this prompt
https://preview.redd.it/v7zv3hnn4u4g1.jpg?width=624&format=pjpg&auto=webp&s=be3cde2286733c5af8f53c84374b73e2baf4de70
create a horror comics with this uploaded public domain character. She finds a door to a alternate dimension and enters into it. put proper dialogues. need 8 panels.
It gave me this one:
Nano Banana 2 Pro
It's good.
Then I asked Gemini to provide a full "text to image" prompt for reuse.
it gave this one:
An 8-panel sequential comic book page, vintage 1950s horror comics style, featuring a blonde female baseball player in a white uniform with a "B" cap (the public domain character "Babe"). Red and green color palette, thick ink outlines, heavy shadows, and Ben-Day dots for texture.
Panel 1: (Normal Reality) "Babe" at bat on a normal baseball field. Dialogue: "Just a practice swing."
Panel 2: (Discovery) "Babe" looking surprised at a mysterious, glowing, old wooden door standing alone in the outfield grass. Dialogue: "What the... A door? Here?"
Panel 3: (Investigation) Close-up of "Babe" reaching for the door, looking puzzled. Dialogue: "No lock... Where does it go?"
Panel 4: (Entry) "Babe" opening the door slightly, looking into darkness. Dialogue: "Hello? Anyone there?"
Panel 5: (Transition) "Babe" standing in a swirling vortex/portal of purple, black, and red. She is mid-step, looking panicked. Dialogue: "Whoa! What's happening?!"
Panel 6: (Alternate Dimension) A strange, surreal, alien baseball field. The ground is purple, the grass is bizarre fungi and glowing flora, with strange, menacing mushrooms surrounding the diamond. The field itself looks cracked and otherworldly. Dialogue: "This isn't the field..."
Panel 7: (Horror Reveal) A monstrous, horrifying, Cthulhu-esque creature bursts from the ground, covered in tentacles and dripping slime, roaring at "Babe." The background is a flash of intense red light. Dialogue: "GAAH! What is that?!"
Panel 8: (Action/Climax) "Babe" winds up to swing her baseball bat at the creature, eyes wide with determination and fear. She is framed against a violent red background, with motion lines around the bat. Onomatopoeia "CRACK!" is written near the impact point. Dialogue: "GET AWAY FROM ME!".
Then, I gave this one to Z-Image Turbo & Flux 2.
They nailed it.
Z-Image:
z-image
Flux 2 dev:
Flux 2
\--
I actually flabbergasted due to those outputs. Because NB 2 Pro is a closed source pipeline. So it produced a complex output.
But Z-Image & Flux 2 dev are open source single models. They able to create a comics.
Z-Image lags behind in text accuracy than Flux 2. But it will improve.
Also note that, since Gemini gave 'text desc' for the character uploaded, the other able to reproduce something near.
I for see a great promising future.
https://redd.it/1pcgqdm
@rStableDiffusion
Quick PSA, the StableDiffusioncpp implementation of z-image is up to 2x faster than the comfyui implementation on some cards.
I was first alerted to this by someone in a discord server and I felt this was important enough to share. There's a good chance that this only applies to 20 series (and maybe some other non-bf16 cards) but if you're interested I suggest giving it a try. This is very informal, I'm just reporting my experience in case it's useful to others.
With my 2060 6GB and the fp16 hack I get about 7.5-8s/it on z-image with comfyui. I've tried several different speedups like cache-dit and the gguf nodes (to limit offloading), but they either look noticeably worse (cache-dit) or make no difference (gguf).
Now with StableDiffusioncpp I'm getting 4s/it, nearly a 2x speed increase without any noticeable quality degradation.
I'm using the newest (at the time of writing) versions of both. Comfy is the portable version while I compiled sdcpp myself (too much work to troubleshoot the precompiled cuda binaries).
I assume the difference comes primarily from more effective use of the gguf format and better offloading. It's also possible that the way sdcpp handles the fp32 upcasting required on 20 series cards is more efficient than what the comfyui hack does.
The big drawback here is obviously the lack of flexibility compared to comfy, but for me at least it's a big enough gain that I'll probably slop together a wrapper node for it at some point (if it doesn't already exist).
Here's the command I'm using for sdcpp:
To get this image:
https://preview.redd.it/um059qe04u4g1.png?width=1024&format=png&auto=webp&s=9c79951152a301aabd5b8b0098fb512cb7cd127a
You might notice that I'm using the heretic version of qwen. This is simply because I already had the gguf version downloaded from when I tested it previously. In practice it makes a very mild difference.
Instructions for setting up z-image in stablediffusioncpp are here: https://github.com/leejet/stable-diffusion.cpp/blob/master/docs/z\_image.md.
Huge props to leejet and everyone else who works on stablediffusioncpp.
https://redd.it/1pchpjb
@rStableDiffusion
I was first alerted to this by someone in a discord server and I felt this was important enough to share. There's a good chance that this only applies to 20 series (and maybe some other non-bf16 cards) but if you're interested I suggest giving it a try. This is very informal, I'm just reporting my experience in case it's useful to others.
With my 2060 6GB and the fp16 hack I get about 7.5-8s/it on z-image with comfyui. I've tried several different speedups like cache-dit and the gguf nodes (to limit offloading), but they either look noticeably worse (cache-dit) or make no difference (gguf).
Now with StableDiffusioncpp I'm getting 4s/it, nearly a 2x speed increase without any noticeable quality degradation.
I'm using the newest (at the time of writing) versions of both. Comfy is the portable version while I compiled sdcpp myself (too much work to troubleshoot the precompiled cuda binaries).
I assume the difference comes primarily from more effective use of the gguf format and better offloading. It's also possible that the way sdcpp handles the fp32 upcasting required on 20 series cards is more efficient than what the comfyui hack does.
The big drawback here is obviously the lack of flexibility compared to comfy, but for me at least it's a big enough gain that I'll probably slop together a wrapper node for it at some point (if it doesn't already exist).
Here's the command I'm using for sdcpp:
.\stable-diffusion.cpp\build\bin\Release\sd.exe --diffusion-model ./z_image_turbo-Q4_K.gguf --vae ./ae.sft --llm ./qwen-4b-zimage-heretic-q8.gguf -p "a photo of a skeleton wearing tophat and a black t-shirt with white text that says '4s/it on a 2060' in a cursive font" --cfg-scale 1.0 -v --offload-to-cpu --diffusion-fa -H 1024 -W 1024 --steps 8To get this image:
https://preview.redd.it/um059qe04u4g1.png?width=1024&format=png&auto=webp&s=9c79951152a301aabd5b8b0098fb512cb7cd127a
You might notice that I'm using the heretic version of qwen. This is simply because I already had the gguf version downloaded from when I tested it previously. In practice it makes a very mild difference.
Instructions for setting up z-image in stablediffusioncpp are here: https://github.com/leejet/stable-diffusion.cpp/blob/master/docs/z\_image.md.
Huge props to leejet and everyone else who works on stablediffusioncpp.
https://redd.it/1pchpjb
@rStableDiffusion