batchbinary/flux-qfaces – Run with an API on Replicate

batchbinary / flux-qfaces

Flux trained on QFACES of High quality facial textures of diverse people, age, ethnicity. Mainly used for image to image (img2img) enhancement. See Readme for ideas & settings.

Warm

Public
2.4K runs
H100

Run with an API

Playground API Examples README Versions

Input

prompt

*string

Shift + Return to add a new line

photo, high resolution, face, high quality, QFACES, photograph, skin, hair, detailed, maximum definition, 4k, highres,photo, high resolution, face, high quality, QFACES, photograph, skin, hair, detailed, maximum definition, 4k, highres,

Prompt for generated image. If you include the `trigger_word` used in the training process you are more likely to activate the trained object, style, or concept in the resulting image.

image

file

Preview

Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored.

prompt_strength

number

(minimum: 0, maximum: 1)

Prompt strength when using img2img. 1.0 corresponds to full destruction of information in image

Default: 0.8

model

string

Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps.

Default: "dev"

num_inference_steps

integer

(minimum: 1, maximum: 50)

Number of denoising steps. More steps can give more detailed images, but take longer.

Default: 28

guidance_scale

number

(minimum: 0, maximum: 10)

Guidance scale for the diffusion process. Lower values can give more realistic images. Good values to try are 2, 2.5, 3 and 3.5

Default: 3

seed

integer

Random seed. Set for reproducible generation

output_format

string

Format of the output images

Default: "webp"

output_quality

integer

(minimum: 0, maximum: 100)

Quality when saving the output images, from 0 to 100. 100 is best quality, 0 is lowest quality. Not relevant for .png outputs

Default: 80

disable_safety_checker

boolean

Disable safety checker for generated images.

Default: false

go_fast

boolean

Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16

Default: false

megapixels

string

Approximate number of megapixels for generated image

Default: "1"

lora_scale

number

(minimum: -1, maximum: 3)

Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora.

Default: 1

extra_lora

string

Shift + Return to add a new line

Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars'

extra_lora_scale

number

(minimum: -1, maximum: 3)

Determines how strongly the extra LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora.

Default: 1

Including mask and 4 more...

mask

file

Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored.

aspect_ratio

string

Aspect ratio for the generated image. If custom is selected, uses height and width below & will run in bf16 mode

Default: "1:1"

height

integer

(minimum: 256, maximum: 1440)

Height of generated image. Only works if `aspect_ratio` is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation

width

integer

(minimum: 256, maximum: 1440)

Width of generated image. Only works if `aspect_ratio` is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation

num_outputs

integer

(minimum: 1, maximum: 4)

Number of outputs to generate

Default: 1

Run this model in Node.js with one line of code:

npx create-replicate --model=batchbinary/flux-qfaces

or set up a project from scratch

Install Replicate’s Node.js client library:

npm install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import and set up the client:

import Replicate from "replicate";

const replicate = new Replicate({
  auth: process.env.REPLICATE_API_TOKEN,
});

Run batchbinary/flux-qfaces using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

const output = await replicate.run(
  "batchbinary/flux-qfaces:4d1d2ea165dffc2d84792d350e9819e3eeb03dc6ceaffcd9c859bbc2d7301f7e",
  {
    input: {
      seed: 42,
      image: "https://replicate.delivery/pbxt/MFoN60GJkZUE5qH6bejQ3yQBG2pXHIIT6nmIF9ObWmCkYEFu/test-A4-CF-thenCR1.png",
      model: "dev",
      width: 1440,
      height: 1440,
      prompt: "photo, high resolution, face, high quality, QFACES, photograph, skin, hair, detailed, maximum definition, 4k, highres,",
      go_fast: false,
      lora_scale: 1,
      megapixels: "1",
      num_outputs: 1,
      aspect_ratio: "custom",
      output_format: "png",
      guidance_scale: 3,
      output_quality: 80,
      prompt_strength: 0.4,
      extra_lora_scale: 1,
      num_inference_steps: 50
    }
  }
);

// To access the file URL:
console.log(output[0].url()); //=> "http://example.com"

// To write the file to disk:
fs.writeFile("my-image.png", output[0]);

To learn more, take a look at the guide on getting started with Node.js.

Install Replicate’s Python client library:

pip install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import the client:

import replicate

Run batchbinary/flux-qfaces using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

output = replicate.run(
    "batchbinary/flux-qfaces:4d1d2ea165dffc2d84792d350e9819e3eeb03dc6ceaffcd9c859bbc2d7301f7e",
    input={
        "seed": 42,
        "image": "https://replicate.delivery/pbxt/MFoN60GJkZUE5qH6bejQ3yQBG2pXHIIT6nmIF9ObWmCkYEFu/test-A4-CF-thenCR1.png",
        "model": "dev",
        "width": 1440,
        "height": 1440,
        "prompt": "photo, high resolution, face, high quality, QFACES, photograph, skin, hair, detailed, maximum definition, 4k, highres,",
        "go_fast": False,
        "lora_scale": 1,
        "megapixels": "1",
        "num_outputs": 1,
        "aspect_ratio": "custom",
        "output_format": "png",
        "guidance_scale": 3,
        "output_quality": 80,
        "prompt_strength": 0.4,
        "extra_lora_scale": 1,
        "num_inference_steps": 50
    }
)
print(output)

To learn more, take a look at the guide on getting started with Python.

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Run batchbinary/flux-qfaces using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

curl -s -X POST \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Prefer: wait" \
  -d $'{
    "version": "batchbinary/flux-qfaces:4d1d2ea165dffc2d84792d350e9819e3eeb03dc6ceaffcd9c859bbc2d7301f7e",
    "input": {
      "seed": 42,
      "image": "https://replicate.delivery/pbxt/MFoN60GJkZUE5qH6bejQ3yQBG2pXHIIT6nmIF9ObWmCkYEFu/test-A4-CF-thenCR1.png",
      "model": "dev",
      "width": 1440,
      "height": 1440,
      "prompt": "photo, high resolution, face, high quality, QFACES, photograph, skin, hair, detailed, maximum definition, 4k, highres,",
      "go_fast": false,
      "lora_scale": 1,
      "megapixels": "1",
      "num_outputs": 1,
      "aspect_ratio": "custom",
      "output_format": "png",
      "guidance_scale": 3,
      "output_quality": 80,
      "prompt_strength": 0.4,
      "extra_lora_scale": 1,
      "num_inference_steps": 50
    }
  }' \
  https://api.replicate.com/v1/predictions

To learn more, take a look at Replicate’s HTTP API reference docs.

Output

{
  "completed_at": "2025-01-02T23:28:02.129519Z",
  "created_at": "2025-01-02T23:27:52.664000Z",
  "data_removed": false,
  "error": null,
  "id": "kxnn4fybb1rm80cm55c92kd580",
  "input": {
    "seed": 42,
    "image": "https://replicate.delivery/pbxt/MFoN60GJkZUE5qH6bejQ3yQBG2pXHIIT6nmIF9ObWmCkYEFu/test-A4-CF-thenCR1.png",
    "model": "dev",
    "width": 1440,
    "height": 1440,
    "prompt": "photo, high resolution, face, high quality, QFACES, photograph, skin, hair, detailed, maximum definition, 4k, highres,",
    "go_fast": false,
    "lora_scale": 1,
    "megapixels": "1",
    "num_outputs": 1,
    "aspect_ratio": "custom",
    "output_format": "png",
    "guidance_scale": 3,
    "output_quality": 80,
    "prompt_strength": 0.4,
    "extra_lora_scale": 1,
    "num_inference_steps": 50
  },
  "logs": "2025-01-02 23:27:53.016 | DEBUG    | fp8.lora_loading:apply_lora_to_model:574 - Extracting keys\n2025-01-02 23:27:53.017 | DEBUG    | fp8.lora_loading:apply_lora_to_model:581 - Keys extracted\nApplying LoRA:   0%|          | 0/304 [00:00<?, ?it/s]\nApplying LoRA:  96%|█████████▌| 292/304 [00:00<00:00, 2913.15it/s]\nApplying LoRA: 100%|██████████| 304/304 [00:00<00:00, 2834.08it/s]\n2025-01-02 23:27:53.124 | DEBUG    | fp8.lora_loading:apply_lora_to_model:574 - Extracting keys\n2025-01-02 23:27:53.125 | DEBUG    | fp8.lora_loading:apply_lora_to_model:581 - Keys extracted\nApplying LoRA:   0%|          | 0/304 [00:00<?, ?it/s]\nApplying LoRA:  50%|█████     | 153/304 [00:00<00:00, 1520.65it/s]\nApplying LoRA: 100%|██████████| 304/304 [00:00<00:00, 1768.12it/s]\n2025-01-02 23:27:53.297 | SUCCESS  | fp8.lora_loading:unload_loras:564 - LoRAs unloaded in 0.28s\n2025-01-02 23:27:53.298 | INFO     | fp8.lora_loading:convert_lora_weights:498 - Loading LoRA weights for /src/weights-cache/11c8b9548dd3adca\n2025-01-02 23:27:53.418 | INFO     | fp8.lora_loading:convert_lora_weights:519 - LoRA weights loaded\n2025-01-02 23:27:53.418 | DEBUG    | fp8.lora_loading:apply_lora_to_model:574 - Extracting keys\n2025-01-02 23:27:53.418 | DEBUG    | fp8.lora_loading:apply_lora_to_model:581 - Keys extracted\nApplying LoRA:   0%|          | 0/304 [00:00<?, ?it/s]\nApplying LoRA:  91%|█████████▏| 278/304 [00:00<00:00, 2777.94it/s]\nApplying LoRA: 100%|██████████| 304/304 [00:00<00:00, 2754.48it/s]\n2025-01-02 23:27:53.529 | SUCCESS  | fp8.lora_loading:load_lora:539 - LoRA applied in 0.23s\nUsing seed: 42\nImage detected - setting to img2img mode\nInput image size: 1536x1800\nScaling image down to 1228x1440\nInput image size set to: 1232x1440\n0it [00:00, ?it/s]\n1it [00:00,  4.64it/s]\n2it [00:00,  3.25it/s]\n3it [00:00,  2.93it/s]\n4it [00:01,  2.81it/s]\n5it [00:01,  2.76it/s]\n6it [00:02,  2.71it/s]\n7it [00:02,  2.70it/s]\n8it [00:02,  2.69it/s]\n9it [00:03,  2.67it/s]\n10it [00:03,  2.67it/s]\n11it [00:03,  2.66it/s]\n12it [00:04,  2.66it/s]\n13it [00:04,  2.66it/s]\n14it [00:05,  2.66it/s]\n15it [00:05,  2.65it/s]\n16it [00:05,  2.65it/s]\n17it [00:06,  2.64it/s]\n18it [00:06,  2.65it/s]\n19it [00:07,  2.65it/s]\n20it [00:07,  2.65it/s]\n20it [00:07,  2.71it/s]\nTotal safe images: 1 out of 1",
  "metrics": {
    "predict_time": 9.255331625,
    "total_time": 9.465519
  },
  "output": [
    "https://replicate.delivery/xezq/frY5jInWfvlKOUvPe5JhZZhOH5sL4ZBqe7g0fYxKnK7UgpJgC/out-0.png"
  ],
  "started_at": "2025-01-02T23:27:52.874187Z",
  "status": "succeeded",
  "urls": {
    "stream": "https://stream.replicate.com/v1/files/bcwr-izbgibw3iepuc6ayndu72d4oeq7pzmbm2afqagbnvyc44viyumna",
    "get": "https://api.replicate.com/v1/predictions/kxnn4fybb1rm80cm55c92kd580",
    "cancel": "https://api.replicate.com/v1/predictions/kxnn4fybb1rm80cm55c92kd580/cancel"
  },
  "version": "4d1d2ea165dffc2d84792d350e9819e3eeb03dc6ceaffcd9c859bbc2d7301f7e"
}

Generated in

9.3 seconds

Tweak it ShareReport

2025-01-02 23:27:53.016 | DEBUG    | fp8.lora_loading:apply_lora_to_model:574 - Extracting keys
2025-01-02 23:27:53.017 | DEBUG    | fp8.lora_loading:apply_lora_to_model:581 - Keys extracted
Applying LoRA:   0%|          | 0/304 [00:00<?, ?it/s]
Applying LoRA:  96%|█████████▌| 292/304 [00:00<00:00, 2913.15it/s]
Applying LoRA: 100%|██████████| 304/304 [00:00<00:00, 2834.08it/s]
2025-01-02 23:27:53.124 | DEBUG    | fp8.lora_loading:apply_lora_to_model:574 - Extracting keys
2025-01-02 23:27:53.125 | DEBUG    | fp8.lora_loading:apply_lora_to_model:581 - Keys extracted
Applying LoRA:   0%|          | 0/304 [00:00<?, ?it/s]
Applying LoRA:  50%|█████     | 153/304 [00:00<00:00, 1520.65it/s]
Applying LoRA: 100%|██████████| 304/304 [00:00<00:00, 1768.12it/s]
2025-01-02 23:27:53.297 | SUCCESS  | fp8.lora_loading:unload_loras:564 - LoRAs unloaded in 0.28s
2025-01-02 23:27:53.298 | INFO     | fp8.lora_loading:convert_lora_weights:498 - Loading LoRA weights for /src/weights-cache/11c8b9548dd3adca
2025-01-02 23:27:53.418 | INFO     | fp8.lora_loading:convert_lora_weights:519 - LoRA weights loaded
2025-01-02 23:27:53.418 | DEBUG    | fp8.lora_loading:apply_lora_to_model:574 - Extracting keys
2025-01-02 23:27:53.418 | DEBUG    | fp8.lora_loading:apply_lora_to_model:581 - Keys extracted
Applying LoRA:   0%|          | 0/304 [00:00<?, ?it/s]
Applying LoRA:  91%|█████████▏| 278/304 [00:00<00:00, 2777.94it/s]
Applying LoRA: 100%|██████████| 304/304 [00:00<00:00, 2754.48it/s]
2025-01-02 23:27:53.529 | SUCCESS  | fp8.lora_loading:load_lora:539 - LoRA applied in 0.23s
Using seed: 42
Image detected - setting to img2img mode
Input image size: 1536x1800
Scaling image down to 1228x1440
Input image size set to: 1232x1440
0it [00:00, ?it/s]
1it [00:00,  4.64it/s]
2it [00:00,  3.25it/s]
3it [00:00,  2.93it/s]
4it [00:01,  2.81it/s]
5it [00:01,  2.76it/s]
6it [00:02,  2.71it/s]
7it [00:02,  2.70it/s]
8it [00:02,  2.69it/s]
9it [00:03,  2.67it/s]
10it [00:03,  2.67it/s]
11it [00:03,  2.66it/s]
12it [00:04,  2.66it/s]
13it [00:04,  2.66it/s]
14it [00:05,  2.66it/s]
15it [00:05,  2.65it/s]
16it [00:05,  2.65it/s]
17it [00:06,  2.64it/s]
18it [00:06,  2.65it/s]
19it [00:07,  2.65it/s]
20it [00:07,  2.65it/s]
20it [00:07,  2.71it/s]
Total safe images: 1 out of 1

Examples

View more examples

Run time and cost

This model costs approximately $0.016 to run on Replicate, or 62 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia H100 GPU hardware. Predictions typically complete within 11 seconds.

Readme

Flux trained on ‘QFACES’ = High Quality Facial textures and photos of diverse people, age, ethnicity. Use for improving existing photos with Flux, while maintaining the core details of the image and people within it (img2img, image to image). Best with images of people close up but can generate content, by adding a content description to the prompt. If you try to upscale an image with multiple people at a distance, then increase the ‘Prompt Strength’ and ‘Guidance Scale’ and add a better description of what the photo contains in the prompt. Remember the higher these values the more the model will generate infill content (i.e. ‘make things up’). If you want to upscale an image of someone you or others can identify ( a family member ) then keep these values low and use a close up picture.
If you are looking to improve the facial textures of a picture of someone you know & recognise, then copy the following settings into the JSON tab. (Yes, you can set the Guidance Scale to 0, this works well to keep the person looking familiar), but prime the model with a good prompt and use the training keyword QFACES in the prompt to get better likeness:

{
“model”: “dev”,
“prompt”: “photo, high resolution, faces, people, high quality, QFACES, photograph, “,
“go_fast”: false,
“lora_scale”: 0.5,
“megapixels”: “1”,
“num_outputs”: 1,
“aspect_ratio”: “1:1”,
“output_format”: “png”,
“guidance_scale”: 0,
“output_quality”: 80,
“prompt_strength”: 0.3,
“extra_lora_scale”: 1,
“num_inference_steps”: 50
}
Note: if you copy paste this JSON code and it has BR commands in (i.e. HTML styling is copied in), you need to remove the BR codes first.
Important: the images used to fine tune (QFACE) this model are all copyright free public domain.