Sign In

Facezoom for Kontext

91

1.4k

50

Updated: Jul 4, 2025

conceptfacezoomkontext

Verified:

SafeTensor

Type

LoRA

Stats

531

0

Reviews

Published

Jul 2, 2025

Base Model

Flux.1 Kontext

Training

Steps: 2,800
Epochs: 1

Usage Tips

Strength: 1.2

Trigger Words

facezoom

Hash

AutoV2
457F68B620
default creator card background decoration
Thisland's Avatar

Thisland

The FLUX.1 [dev] Model is licensed by Black Forest Labs. Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs. Inc.

IN NO EVENT SHALL BLACK FOREST LABS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

Facezoom

Boom Headshot! Zooms into a face and enhances detail. Works much better than vanilla kontext.
In comparison images, the cropped original is an auto bbox crop with yolo which is something you can use but this lora could probably replace a face crop + upscale in your workflows.

V1.1:

Trained at 1024px vs 768px
Produces more consistent crops
Better details

Less face elongation

Prompt

f4cezoom

Can add other face modifications, expressions, eye color etc as modifiers. Anything that kontext can do as standard.

Flux guidance: 3 (2.5-3.5)

Denoise: 1 (0.95-1)

Strength: 1.25 (1.00-1.25)


Limitations

  • It's not magic, heavily pixelated or noisy images won't work well

  • Seed variability, if it doesn't get the correct crop you want try a different seed

  • Sometimes faces may be slightly elongated

  • Doesn't work as well for landscape orientation

  • Mixed results with multiple subjects

Trained with Ostris AI-Toolkit

Settings

      train_unet: true

      train_text_encoder: false

      gradient_checkpointing: true

      noise_scheduler: flowmatch

      optimizer: adamw8bit

      timestep_type: sigmoid

      content_or_style: balanced

      optimizer_params:

        weight_decay: 0.0001

      unload_text_encoder: true

      lr: 0.0001

      ema_config:

        use_ema: true

        ema_decay: 0.99