Image LoRA Guide (for Text to Image & Reference-To-Image & Image Editor)

What is a LoRA?

LoRA (Low-Rank Adaptation) is a compact add-on trained to nudge a base model toward a specific look, body shape, style, or concept — without retraining the whole model.

ZenCreator offers two different kinds of LoRAs, and they work in very different ways:

Part 1: SDXL Body-Shape LoRAs

Why use LoRAs?

• Precision control over physique, age cues, or stylistic traits.

• Consistency across a batch or series (e.g., keep the same body shape across outfits).

• Modularity: mix different, compatible traits without hard-coding them into the prompt.

How LoRA strength works

• Range:

  • 0.1 → 3.0 per LoRA (default 0.8).

  • 0.1–0.4 = subtle hint;

  • 0.5–1.2 = balanced control (recommended);

  • >1.5 = aggressive and likely to introduce artefacts (warping, seams, texture glitches).

• When stacking multiple LoRAs, lower individual strengths (e.g., 0.6–0.9 each).

• If you see artefacts, drop a LoRA or reduce strengths in 0.1–0.2 steps and simplify the prompt.

Using multiple LoRAs

• You can apply up to 3 at once.

• Prefer complementary combinations (e.g., Soft Fuller Figure + Thick Thighs & Wide Hips).

• Avoid contradictory pairs at high strength (e.g., Slim Figure + Plus Size Body). If you must combine opposites, keep each ≤0.6 and let the prompt decide the balance.

• Order doesn’t matter; strengths do.

Provider choice (quick rule)

• WAN: most photoreal and detailed; no LoRA support.

• SDXL: flexible and supports LoRAs. Choose SDXL whenever you need body/shape/style control.

LoRA catalog (intent & usage)

Each line explains what the LoRA is for, when to use it, and any notes. The aliases in parentheses map to your internal packs, so your team knows the source.

Large Breast & Hourglass

Emphasises fuller breasts with a classic hourglass silhouette. Great for “bombshell” aesthetics.

Adjustable Large Breast

Breast emphasis with a different learned bias than v2; use when the first option under- or overshoots.

Adjustable Large Breast 2

Alternative slider trained specifically for SDXL; try this if the above conflicts with outfit/pose.

Cameltoe

Enforces tight garment fabric tension in the groin area (swimwear/activewear shots).

Soft Fuller Figure

Adds overall softness and fuller fat distribution. Natural look when combined with lifestyle prompts.

Thick Thighs & Wide Hips

Stronger lower-body emphasis and wider hip line.

Plus Size Body

Plus-size proportions across torso and limbs.

Soft Tummy & Curves

Visible belly softness and gentle curves.

Elegant Mature

Mature facial cues and styling.

Athletic Tone

Toned physique with low-to-moderate body fat.

Strong Muscular Body

Clearly muscular build and definition.

Slim Figure

Slender proportions with minimal body fat.

Classic Hourglass Shape

Tight waist with balanced bust/hip line.

Triple Breast

Fantasy/anatomical variation for creative shoots.

Best-practice workflow with LoRAs

1. Write the core prompt (subject, camera, lighting, mood).

2. Pick SDXL and add 1–2 LoRAs at 0.7–0.9 each.

3. Add a negative prompt for anatomy issues (e.g., “extra limbs, deformed, bad hands, distorted proportions, watermark”).

4. Generate 4–8 images, review artefacts, and adjust strengths by ±0.2.

5. When satisfied, run larger batches.

6. Finalise with Upscaler (for detail) and Head & Face Swap last if you need a specific identity.

Prompting tips with LoRAs

• Keep trait words in the positive prompt so the LoRA has context (e.g., “editorial fashion, hourglass figure”).

• Don’t fight your LoRA with contradictory text (“slim figure” + Plus Size Body).

• Use clean lighting and camera terms; messy prompts amplify artefacts.

Can you combine LoRAs?

Yes — up to three. Combine complementary traits and keep strengths moderate. If the result looks “pulled in different directions,” remove one LoRA or let a single LoRA lead at ~0.9 while the others sit at ~0.4–0.6.

Quick troubleshooting

• Anatomy glitches / seams → lower strengths; remove a conflicting LoRA; simplify the prompt.

• Look not strong enough → raise one LoRA gradually (e.g., 0.8 → 1.0) rather than stacking more.

• Soft/low detail → route to Upscaler.

• Identity mismatch → finish with Head & Face Swap as the final step.

Example starter settings

• Lifestyle hourglass: SDXL + Classic Hourglass Shape 0.8 + Large Breast & Hourglass 0.6.

• Athletic editorial: SDXL + Athletic Tone 0.9 + Slim Figure 0.6.

• Plus-size fashion: SDXL + Plus Size Body 0.9 + Thick Thighs & Wide Hips 0.7.

Part 2: Flux Klein Style LoRAs

Flux Klein LoRAs work completely differently from SDXL body-shape LoRAs. Instead of nudging a text-to-image generation, they're style templates: you upload your own photo, pick a Flux Klein LoRA, and the tool transforms your photo to match that template's look — pose, styling, and overall aesthetic.

How it works

1. Open Image Editor and select the Flux Klein LoRA model.

2. Upload the photo you want to transform.

3. Pick a LoRA style template from the list.

4. Generate — the result follows the template's look while carrying over your subject.

How this differs from SDXL Body-Shape LoRAs

Available Flux Klein LoRA templates

[TODO for team: list the current template catalog here — name + one-line description of what each style does, once the visual examples are ready]

Tips