š FameGrid ā Z-Image Base Checkpoint (Flagship Release)
This checkpoint is built on Z-Image Base and is focused on producing modern, social-media-style photography with a more natural overall look.
The goal wasnāt to push a heavy aestheticāit was to get consistent results that feel believable while still holding good composition and lighting.
USE THE NEWEST HYBRID WORKFLOW for similar results.
š Join UltraMuse Discord (Additional Workflows, Support, Resources, Extra Checkpoints)
What It Does
The model leans toward:
natural-looking skin with some texture (not overly smoothed or plastic)
clean, modern lighting similar to current social content
simple but effective compositions
subjects that feel realistic, not overly āmodel-likeā
Itās designed to sit in that middle ground where images look polished but still pass as real.
š§° Built For
Influencer-style photography
Lifestyle content
Product + human shots
Scalable character generation
If youāre trying to create images that actually blend into real feedsāthis is the tool.
UltraMuse Discord is where the real stuff lives:
ā
Additional workflows (including v1 + updates)
ā
Help & support (prompt + settings troubleshooting)
ā
Other resources (tutorials, experiments, presets)
š Join UltraMuse Discord (Official Workflows, Support, Resources, Extra Checkpoints)
Description
FAQ
Comments (22)
models now are so good, it transcends the reality šā¤ļø
I have to admit, sometimes I lose touch with reality and doubt myself because there are so many styles to these realistic generations. The only ones that snap me out of it are the plain candid generations.
Is it possible to use the alibaba-pai/Z-Image-Fun-Lora-Distill (Distill LoRA for Z-Image that distills both steps and CFG) with this great checkpoint. Also huge congratulations for making a very, very good looking high quality checkpoint for the base version as it definitely seems harder to tame compared to the turbo variant. I do love the diverse outputs he base version gives you though. I'm a huge fan of your work so well done.
I haven't tried on this checkpoint but I did on the base.. It tends to lose a quite a bit of quality, though it should still work.
What are the recommending settings? I'm not using comfyui. I'm using forge neo. Steps, cfg, schedulers, sampler, hires fix, denoise?
when in doubt, use the vanilla z-image base parameters. its like 3-5 CFG, 30-50 steps, euler/beta or dpm2pp_sde/beta (or sgm uniform instead of beta) -No hi res fix needed, 1.00 denoise (default)
So I noticed that this says "checkpoint trained" instead of "checkpoint merged" like I see with other ZIT checkpoints others have made. does that mean I would be able to successfully train a ZIT character from this?
checkpoint merged means someone mixed multiple checkpoints together, rather than training one. Both types you can train a lora on.
Are you planning to release a quantized version? Iāve got an RTX 2060 with 12 GB of VRAM, and Iām not sure if your checkpoint would run on it. Am I wrong?
I 1up this. A small model of base is nice to get a unique image that can be refined in a turbo model.
@Loraholic @DeckJaniels
I converted the model to fp8. The workflow the author uses is not any I could find and I cant get access to the discord to ask for it so good luck getting matching results from the previews.
https://huggingface.co/ApacheOne/ZImageBase-nvfp4_mixed/tree/main
Ill look into it. But right now I'm using it with 12gb ram with no issues.
@KSAPE heya the workflow is on pastebin as well
https://pastebin.com/AL7N2JzX
@UltraMuseĀ seems z image base doesn't handle fp8 too well anyway
Do you prefer ZIB over ZIT? I've really liked the verity you get on the base model.
ZIB definitely creates much better compositions, so that's the one I prefer
could you make a klein checkpoint/lora?
What's the recommended scheduler for res multistep on this ?
this checkpoint is the best I've seen so far for z-image base
Fantastic checkpoint for sadomasochist. If you are willing to suffer, struggle, waiting endless minutes for broken generation thats top tier