After the success of the Rebel Essence LoRA, the next step was inevitable: turning that aesthetic into a full, standalone checkpoint. This is how JRD_Rebel Checkpoint was born — built from scratch on the Z‑Image Turbo pipeline and powered by the Qwen3 + Lumina 2 encoder combo for a truly photographic, physical and cinematic output.
The model is exclusively female. It handles mixed ethnicities, realistic skin, complex materials and strong punk, gothic and dark styling with a stability no LoRA can match. No triggers, no keywords required: the style is embedded directly into the weights.
Rebel Essence is no longer just an aesthetic. It’s a complete model.
Description
FAQ
Comments (14)
The time it takes to generate the image on 11GB is terrible, but I couldn't resist :D
-------
It's great but I'm waiting for a smaller version:D
Lume Blender has been removed, I'll probably remove Rebel Checkpoint too, they don't deserve all this work I do
@JRD_2212 Well I'd love to leave a comment about how awesome this is, but I haven't been able to try it since I'm like the other guy above who doesn't have the latest video card.
@ferrrett33 which model are you talking about exactly...maybe I'll send it to you via google drive link, very willingly
12GB cp model v1.0 literally generates 1280x1600 images in 45 secs with 7 steps and 1.0 cfg. Use ModelComputeDtype node and put fp16 there and run CLIP encoder on CPU so there is no issue to run this on old hardware with 8GB VRAM (have 32GB RAM though to cache things).
About checkpoint itself... gaping vagina or wide open vagina needs a bit more work.
@JRD_2212 Yes this one, thanks for the FP8s, I've added your models so will be testing them in my list for the next few months. Thanks bro
Could you please provide your workflow? I am getting this error when I am using this checkpoint in my normal Z-Turbo Comfy workflows. !!! Exception during processing !!! shape '[10240, 3840]' is invalid for input of size 10231816
You’re loading the wrong CLIP Text Encoder.
This is a Z-Image Turbo checkpoint, so you must load Qwen 3.4B as the text encoder — not Stable Diffusion / SDXL CLIP.
The 10240 shape error means your workflow is forcing an SDXL encoder, which is incompatible with this model.
Remove any nodes that override the text encoder and load Qwen 3.4B only.
@charidot113 https://huggingface.co/Comfy-Org/z_image_turbo/blob/main/split_files/text_encoders/qwen_3_4b.safetensors
https://huggingface.co/Comfy-Org/z_image_turbo/blob/main/split_files/vae/ae.safetensors
JRD_2212 brilliant but sometimes his translation is mediocre :D Here are the files you need to run his work
Stable Diffusion WebUI Forge is a highly optimized platform designed to generate high-resolution images much faster than standard software. It uses intelligent memory management that loads and removes models dynamically, completely preventing "Out of Memory" crashes on 8 GB VRAM cards. By halving the system load, it allows modest hardware to smoothly run heavy next-gen architectures like FLUX and Wan 2.2. Its computational power relies on advanced token acceleration libraries that boost rendering speeds by 2× to 3×. Finally, it natively supports compressed FP8 and GGUF models, which drastically drops system temperatures and cuts down upscaling times.
https://huggingface.co/OreX/Automatic1111/tree/main/Forge-Neo-v2
If I have any more questions, just ask, although I recommend AI, it helps me the best :D
Thank you both for your advice - much appreciated. In fact I am already using Qwen_3_4b.safetensors (with type lumina2) as the text encoder and ae.safetensors as the VAE. All other Z-Image Turbo models are working fine in the two ComfyUI workflows (including the default template from ComfyUI) that I am using, so I am not quite sure why Rebel is failing.
@PrincesOfDarkness Good point - LLMs might be useful assistants!
@charidot113 if you use discord join my discord server and I'll give you the workflow that I use with checkpoint rebel, and you'll see that you find the solution
my discord channel here: https://discord.gg/yCctrYhk
Problem is now resolved, - my first download of the checkpoint was seemingly incomplete. Second download is now working perfectly.
I think JRD Rebel Checkpoint is perfect. You should take a breather and let people enjoy this Checkpoint, it's brilliant.
It has everything
And to those of you who download it without a word because it's free, I suggest you appreciate his work.
Because if he's gone, we'll lose some very good Checkpoints.
I can't believe you don't like the model.
I'm currently using the FP8 model. It's a lifesaver for my computer.