Recommended settings

• Base model: FLUX.1 Kontext [dev]

• LoRA strength: 0.6–1.0

• Starting strength: 1.0

• Steps: 24–30

• Guidance: 2.5–3.5

• Use the source image’s original aspect ratio whenever possible.

Lower the LoRA strength or guidance if the result becomes too extreme, or if the edit begins changing too many non-target details.

Trigger words:

jbusty

Example Prompt:

jbusty,
change her breasts to be larger while preserving everything else

Notes

The training set focuses on a single localized body edit: increasing breast size while preserving the rest of the image as closely as possible.

It was trained on paired source/target images where the main intended difference is bust size, so it tends to work best when the source image has:

• a clearly visible subject,

• a readable chest silhouette,

• relatively uncomplicated poses,

• and minimal obstruction around the torso.

Results depend heavily on the source image. Straightforward images with clean lighting, clear anatomy, and less visual clutter generally work best.

Because the training set includes a variety of clothing styles, poses, and angles, the LoRA may generalize reasonably well across many image types, but performance will still vary depending on perspective, garment structure, and how visible the chest area is in the source image.

Why this LoRA exists

FLUX Kontext is very strong at many image edits, but in my testing it could be inconsistent with breast enlargement edits. It would sometimes:

• underdo the requested change,

• overdo the effect in an unrealistic way,

• alter unrelated body proportions,

• change clothing more than necessary,

• or drift too far from the original face, pose, lighting, or scene.

JBusty was trained to make bust enlargement edits more reliable and controllable, helping Kontext better preserve the original subject and image while applying the requested body change more consistently.

Simpler than manual reshaping workflows

Another reason for this LoRA is practicality. Many body-editing workflows can be awkward, fragile, or inconsistent, often requiring manual liquify steps, multiple passes, masking, or repeated trial and error.

JBusty is meant to be a simpler alternative:
load the LoRA into a standard FLUX Kontext workflow, prompt the edit directly, and adjust LoRA strength as needed. No complex multi-stage workflow is required.

Future versions

Future versions are planned to improve JBusty with:

• better control over subtle vs stronger enlargement,

• better preservation of body proportions outside the chest area,

• improved clothing refitting and fabric behavior,

• stronger consistency across difficult angles and poses,

• better handling of fitted tops, dresses, swimwear, and layered clothing,

• improved realism and integration under more lighting conditions and camera perspectives.

The goal is to make each update more flexible and more predictable while keeping the workflow simple: load the LoRA, describe the desired enlargement, and run the edit through a standard FLUX Kontext workflow.

Responsible use

This LoRA is based on FLUX.1 Kontext [dev]. Users are responsible for following the applicable FLUX.1 [dev] license, Black Forest Labs Acceptable Use Policy, platform rules, and local laws.

Use only with images you own or have permission to edit. Do not use this LoRA to create anything illegal, deceptive, or non-consensual. JBusty is intended for consensual adult image editing, cosmetic-surgery-style simulation, and creative experimentation.

This model is an independently created derivative and is not affiliated with, sponsored by, or endorsed by Black Forest Labs. The LoRA itself should be treated as non-commercial unless the user has obtained the appropriate commercial rights from Black Forest Labs. FLUX.1 Kontext [dev] is explicitly included under the FLUX.1 [dev] non-commercial license.