🚀 [NEW] Z-Image Revolution ZIB Base

Fine-tuned on a meticulously curated dataset of modern social-media influencer photography (INSTAGRAM STYLE). This LoRA took months of curation, iteration, and an unreasonable amount of server burn to get natural skin, modern lighting, and clean composition to land consistently.

New Hybrid workflow for Revolution
Settings to experiment with:

• Lora weights. If using with a character Lora I suggest making the combined lora weights = 1

• Very Low CFG 1-2 on first Ksampler can get ultra realistic results.

• Denoise on second ksampler 0.2-0.6 (set it to your liking, effects skin quality but can make model features look less realistic)

• Model Shift (higher values tend to give less artifacts but at cost of nice compositions)
  

🆕 FameGrid Revolution — Z-Image Base + Turbo Hybrid (Newest Release)

This is the new flagship.

What it is

• Highest quality yet — trained on the Z-Image Base model.

• Compatible with BOTH: Z-Image Base and Z-Image Turbo.

• Use the Hybrid workflow for best quality (this is the intended way to run it).

About the samples + checkpoints

• Sample images were generated using multiple checkpoints — top 5 from 20 training runs.

• Each checkpoint has its own traits, but quality is generally consistent across the top set.

• The release checkpoint was chosen for best adherence + overall balance.

• Some alternate checkpoints can produce:

  • better environments

  • better skin quality

  • or even better-looking subjects
    …but may trade off adherence/consistency.

More checkpoints (Discord)
Additional checkpoints are available in Discord:

• Get 2 extra (free) or purchase the full pack.
  👉 https://discord.gg/4zbGm5j6jW

UltraMuse Discord is where the real stuff lives:

✅ Official workflows (including v1 + updates)
✅ Help & support (prompt + settings troubleshooting)
✅ Other resources (tutorials, experiments, presets)
✅ Additional checkpoints (stronger and lite) for Z-Image v1 available (not all will hit CivitAI)

👉 Join UltraMuse Discord (Official Workflows, Support, Resources, Extra Checkpoints)

⭐ Additional Checkpoints (What Each One Does)

These are included as optional variants:

FameGrid_Revolution_ZIB_Bold

“Stronger baked-in look”

• More aggressive influencer styling

• Better-looking subjects more often

• Great if you want instant IG polish without fighting prompts

FameGrid_Revolution_ZIB_More_Bold

“Even more IG”

• The strongest “Instagram look” variant

• Heavier aesthetic push / more stylized strength

• Best when you want max influence and don’t care about subtle realism as much

FameGrid_Revolution_ZIB_ExtraReal

“Cleaner realism bias”

• Nicer backgrounds overall

• Subjects lean more average / believable (less “model-face”)

• Great for authentic lifestyle + grounded social content

FameGrid_Revolution_ZIB_UltraReal

“The quality monster”

• Extremely nice backgrounds

• Great prompt adherence

• Minimal artifacts

• Best for “this needs to pass as real” outputs and complex prompts
  

Recommended LoRA Weight (Z-Image v1)

0.7 – 0.9 is the sweet spot

• 0.7 = more natural, less overcooked

• 0.9 = stronger stylization, still clean

🔥 Z-Image v1 PRO (Discord-Only)

PRO version is coming in the next few days:

• It will NOT be released on CivitAI.

• Discord is the only place you’ll get the notification.

👉 Join UltraMuse Discord if you actually want access.

🆕 1.5 Latest Updates (Qwen)

New Trigger Word: rlskn
Use this trigger along with the default one IGMODEL to unlock:

• more realistic skin texture

• natural imperfections

• more “average-looking” believable subjects (less plastic model-face)

Start prompts like:
“iPhone photo of an IGMODEL woman with rlskn (detailed skin:1.3) …”

Additional Guidance

• rlskn = realism, texture, believable skin

• Lower CFG = more realism

• Higher CFG = sharper, but can reduce naturalness

• Match CFG and LoRA weight:

  • Low CFG → lower weights

  • High CFG → higher weights

• This model loves long prompts — 800+ characters is ideal

• If outputs aren’t realistic enough, increase LoRA strength slightly

Who This Is For

FameGrid was trained on imagery emphasizing:

• Natural, confident posing

• Clean, modern composition

• Soft but controlled lighting

• Professional yet relatable influencer aesthetics

Ideal for:

• Lifestyle & influencer-style photography

• Product shots with human context

• Virtual character creation

• Social content generation at scale

Trigger Words

• IGMODEL — main trigger for influencer-style realism

• rlskn — enhanced skin texture + natural appearance

Recommended Settings (Qwen-Image)

• Model Flow: 2

• Sampler: res 2s

• Scheduler: simple

• CFG: 1–3 (1 recommended)

• LoRA Weight: 1.0–2.0 (1 recommended; increase if realism drops)

🧪 Behind the Scenes — The Training Gauntlet

FameGrid didn’t come together cleanly. It was a brutal training cycle — collapses, overfit runs, underfit runs, entire experiments trashed because one stylistic artifact poisoned the look.

I destroyed more models than I kept. Re-curated images, rebuilt datasets, restarted pipelines, burned cloud credits like they were nothing — because that’s what it costs to chase realism without drifting into uncanny, plastic, or over-processed territory.

Every iteration forced one choice: throw it out or settle.
I didn’t settle.

The result is a LoRA that feels natural, modern, and clean — influencer photography that can actually pass in real social environments.

🛠 Additional Notes & Tools

FameGrid is highly sensitive to prompt structure. Tiny phrasing changes can swing results wildly — that’s why it’s powerful, but also why lazy prompts don’t get premium outputs.

To fix that, we built:

FameGrid Prompt Expander
Turns short prompts into long-form descriptions that match the model’s ideal length and composition.

👉 Join UltraMuse Discord (Official Workflows, Support, Resources, Extra Checkpoints)