What the tool does
Generation by Reference analyzes your image (subject, pose, camera angle, lighting, color grade, environment) and produces similar, not identical photos. It keeps the feel of the reference — location, pose family, styling — while exploring natural variations so you can pick the strongest frame.
Great for: recreating a look in different takes, matching a feed’s aesthetic, building sets for carousels, and preparing shots for Face Swap to lock identity.
Interface tour
Upload Reference Images
Drop PNG/JPEG files (up to 1000). Use clean, well-lit photos with a clear subject.Reference Strength
Controls how closely results follow the reference.
Start around 0.8; lower for freer variations, higher for closer matches.Positive Prompt (optional)
Add details you want (e.g., “soft window light, 50mm, neutral grade”). Keep it concise.Negative Prompt (optional)
Ban common issues (e.g., “bad anatomy, extra fingers, watermark, oversharpened”).LoRA (optional)
Attach up to 3 LoRAs to guide physique or style. Begin at 0.6–0.9 strength each; higher values increase artefacts.Comment (optional)
Internal notes for you or teammates.Start Generation
Launches the batch. The per-image credit cost appears next to the button.
Quick start
Upload a clear reference image.
Set Reference Strength to 0.8 (adjust later if needed).
(Optional) Pick up to 3 LoRAs and keep strengths moderate (0.6–0.9).
Add brief Positive/Negative Prompts if you want tighter control.
Click Start Generation.
Open the finished task, select the winners, then Download or Send to another tool.
Working with results
Use the checkboxes on thumbnails to Select/Deselect.
Download Selected / Download All to export a ZIP.
Send to another tool to continue the pipeline (Upscaler, Face Swapping, Generator by Ref, Video Generator, Carousel).
Retry Failed appears automatically if any frames mis-rendered.
Speed & quality flows
Fastest: Upscale + Face Swap flow — runs both in one pass on your selections for the cleanest final look.
Fully automated: Generation by Ref + Upscale + Face Swap flow — generates, upscales, and swaps faces automatically so every output is ready to publish with no tool hopping.
Best practice: Face Swap is most reliable as the final step. If the face must remain unchanged while enlarging, use Face-Safe Upscale.
Pro tips
Quality in → quality out. Choose references with sharp faces, simple backgrounds, and consistent light.
Batch for choice. Generate 8–10 variations per reference to find a hero frame.
Use LoRAs for physique/style. You don’t need a reference with a perfectly consistent body — SDXL + LoRAs will guide it.
FAQ
Do I need physique-consistent references?
No. Use SDXL with LoRAs to shape body type; the reference anchors pose/scene/mood.
How close will the results be to my reference?
Tune Reference Strength: 0.6–0.8 for creative range, 0.85–0.95 for close adherence.
Can I use LoRAs here?
Yes, up to 3 at once. Keep each at 0.6–0.9 to minimize artefacts.
What if the face isn’t the one I need?
Run Face Swap as the final step, or use the combined flow so images come out upscaled and face-swapped automatically.
Is the license commercial?
Yes — outputs are royalty-free for commercial use.
Troubleshooting
Identity feels off vs. reference: raise Reference Strength slightly or Face Swap the selected frames.
Anatomy/clothing glitches: lower LoRA strengths by 0.2, simplify prompts, re-run.
Soft/“plastic” detail: route selects to Upscaler.
Batch inconsistency: lock aspect ratio and seeds; avoid changing lighting/camera terms mid-series.
Need help choosing the right flow for your shoot? Ping the chat bubble or email [email protected].