AI Generator by Prompt (Text-To-Image)

Write a prompt, pick a provider, optionally add LoRAs, and generate a batch you can immediately upscale, face-swap, or publish.

ZenCreator Team
Updated November 24, 2025
5 min read
text-to-imagegenerationlorapromptingwansdxl

What the tool does

Generator by Prompt creates images directly from text. You control subject, setting, lighting, composition, and style. Choose a model provider and (with SDXL) attach LoRAs to steer body type or stylistic details. Results are commercially licensed and can be sent downstream to any ZenCreator tool.

Interface tour

  • Prompt — describe the scene you want. Be specific about subject, pose, camera angle, lighting, mood, and environment. See a Full Guide "How to create a good prompt".
  • Negative Prompt — list things to avoid (extra fingers, blur, bad anatomy, watermark, low contrast…).
  • Number of Images — total outputs for this run.
  • Model — choose WAN or SDXL (details below).
  • LoRA — SDXL only. Pick up to 3 LoRAs and set Strength for each (default 0.8, range 0.1–3.0).
  • Comment (optional) — internal notes for you/teammates.
  • Start Generation — launches the job; the per-image cost is shown next to the button.

Providers: WAN vs. SDXL

  • WAN
    • Best for highly realistic, detailed images.
    • Strong prompt following.
    • Does not support LoRAs.
  • SDXL
    • Versatile base model.
    • Supports LoRAs for body/shape/style control.
    • Ideal when you need a consistent physique or niche attributes.

Quick rule of thumb: If you don’t need LoRAs, start with WAN for realism. If you need shape/style control, pick SDXL + LoRAs.

LoRAs (SDXL)

Attach up to three LoRAs at once and set a Strength per LoRA.

  • 0.1–0.4 = subtle influence
  • 0.5–1.2 = balanced control (recommended)
  • >1.5 = aggressive; more likely to introduce artefacts

Important: Higher strength increases the chance of artefacts. Combine fewer LoRAs at moderate strengths for the cleanest results.

Available LoRAs:

  • Large Breast & Hourglass — fuller bust with classic hourglass balance.
  • Adjustable Large Breast — scalable bust emphasis while keeping overall proportions.
  • Adjustable Large Breast 2 — alternative profile with a different aesthetic bias; useful if v1 conflicts with your scene.
  • Cameltoe — swimwear/activewear fabric tension at groin area.
  • Soft Fuller Figure — soft curves and slightly higher body fat distribution.
  • Thick Thighs & Wide Hips — stronger lower-body emphasis with 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 features and styling cues.
  • Athletic Tone — toned musculature with low-to-moderate body fat.
  • Strong Muscular Body — visibly muscular physique; pronounced definition.
  • Slim Figure — slender proportions with minimal body fat.
  • Classic Hourglass Shape — defined waist with balanced bust/hip ratio.
  • Triple Breast — fantasy/anatomical variation.

See our LoRA usage guide for examples and best practices.

Quick start

  1. Write a precise Prompt (and Negative Prompt for must-avoid items).
  2. Pick a Model: WAN (no LoRAs) or SDXL (with LoRAs).
  3. If using SDXL, you can add up to three LoRAs and set Strength (start at 0.8).
  4. Set the Number of Images and click Start Generation.
  5. Open the finished task, review, Download or Send to another tool (Upscaler, Face Swap, Generator by Ref, Video, Carousel).

Actioning results

  • Select thumbnails to download only the winners.
  • Use Send to another tool to pipeline instantly.
  • If a few outputs fail, use Retry Failed—only missing images are re-rendered.

Prompting tips (copy-paste friendly)

  • Portrait: “35mm portrait of [subject], half-body, eye contact, soft window light, shallow depth of field, natural skin, neutral color grade, realistic texture, high detail” Negative: “lowres, over-smooth skin, bad anatomy, extra fingers, watermark”
  • Lifestyle: “[subject] walking in [location], candid pose, golden hour backlight, film look, natural grain, realistic proportions, composition rule of thirds” Negative: “cartoonish, plastic skin, lens distortion, duplicate face”
  • Fashion: “[subject] studio fashion shot, 3/4 pose, key light + rim, clean backdrop, editorial style, crisp shadows, high-end retouch look” Negative: “blown highlights, muddy blacks, JPEG artefacts”

See a Full Guide "How to create a good prompt".

Pro tips

  • With multiple LoRAs, keep each 0.5–0.9; push above 1.2 only if you must.
  • If outputs look soft or under-detailed, Upscale it.
  • If face identity isn’t what you need, route selects to Face Swap as the final step.

FAQ

How many LoRAs can I use at once? Up to three (SDXL only).

Which provider is more realistic? WAN generally yields the most photoreal look, but SDXL + tuned LoRAs can match it with the right settings.

Why do I see artefacts? Excessive LoRA strength or too many competing LoRAs. Lower strengths, remove one, or simplify the prompt.

Troubleshooting

  • Anatomy or clothing distortions → lower LoRA strengths (e.g., from 1.2 to 0.7) and simplify the prompt.
  • Style not sticking → increase one LoRA slightly (e.g., 0.8 → 1.0) instead of stacking more LoRAs.
  • Over-smooth or low-detail → send to Upscaler.
  • Identity not matching your brief → finish with Face Swap using your source face.