bfl/flux-general/differential-diffusion
Flux General Differential Diffusion is BFL's intelligent image editing model. Transform, retouch, and reimagine existing images using text prompts - from background replacement to artistic style conversion.
Input
The prompt to generate an image from.
PNG, JPEG, WebP, or GIF · 20 MiB maximum
URL of image to use as initial image.
This field is required.
0.011
The strength to use for differential diffusion. 1.0 is completely remakes the image while 0.0 preserves the original. Range: 0.01 to 1.
The format of the generated image. Allowed values: jpeg, png.
05
The CFG (Classifier Free Guidance) scale is a measure of how close you want
the model to stick to your prompt when looking for a related image to show you. Range: 0 to 5.
The proportion of steps to apply NAG. After the specified proportion
of steps has been iterated, the remaining steps will use original
attention processors in FLUX. Maximum: 1.
URL of change map.
Specifies whether beta sigmas ought to be used.
-33
Strength of reference_only generation. Only used if a reference image is provided. Range: -3 to 3.
The tau for NAG. Controls the normalization of the hidden state.
Higher values will result in a less aggressive normalization,
but may also lead to unexpected changes with respect to the original image.
Not recommended to change this value.
0.015
Max shift for the scheduled timesteps Range: 0.01 to 5.
The alpha value for NAG. This value is used as a final weighting
factor for steering the normalized guidance (positive and negative prompts)
in the direction of the positive prompt. Higher values will result in less
steering on the normalized guidance where lower values will result in
considering the positive prompt guidance more. Maximum: 1.
0.015
Base shift for the scheduled timesteps Range: 0.01 to 5.
The scale for NAG. Higher values will result in a image that is more distant
to the negative prompt. Maximum: 10.
PNG, JPEG, WebP, or GIF · 20 MiB maximum
URL of Image for Reference-Only
The LoRAs to use for the image generation which use a control image. You can use any number of LoRAs
and they will be merged together to generate the final image.
Uses classical CFG as in SD1.5, SDXL, etc. Increases generation times and price when set to be true.
If using XLabs IP-Adapter v1, this will be turned on!.
Use an image input to influence the generation. Can be used to fill images in masked areas.
Sigmas schedule for the denoising process.
01
The percentage of the total timesteps when the reference guidance is to be ended. Range: 0 to 1.
IP-Adapter to use for image generation.
EasyControl Inputs to use for image generation.
01
The percentage of the total timesteps when the reference guidance is to bestarted. Range: 0 to 1.
The aspect ratio of the generated image. Allowed values: 21:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16.
The same seed and the same prompt given to the same version of the model
will output the same image every time.
Enter a JSON array.
020
Range: 0 to 20.
150
Range: 1 to 50.
Idle
Example output — click Run to generate your own
Production endpoint
Send your first request
OpenAI-compatible endpoint with unified authentication and usage tracking.
POST
https://api.sandbase.ai/v1/runModel ID
bfl/flux-general/differential-diffusion01
Input Schema
28 parameters · 2 required · 26 optional
| Parameter | Type | Required | Description |
|---|---|---|---|
image | string | Required | URL of image to use as initial image. |
prompt | string | Required | The prompt to generate an image from. |
seed | integer | Optional | The same seed and the same prompt given to the same version of the model will output the same image every time. |
loras | array | Optional | Default: [] |
nag_end | number | Optional | The proportion of steps to apply NAG. After the specified proportion of steps has been iterated, the remaining steps will use original attention processors in FLUX. · Max: 1 · Default: 0.25 |
nag_tau | number | Optional | The tau for NAG. Controls the normalization of the hidden state. Higher values will result in a less aggressive normalization, but may also lead to unexpected changes with respect to the original image. Not recommended to change this value. · Default: 2.5 |
strength | number | Optional | The strength to use for differential diffusion. 1.0 is completely remakes the image while 0.0 preserves the original. · Min: 0.01 · Max: 1 · Default: 0.85 |
max_shift | number | Optional | Max shift for the scheduled timesteps · Min: 0.01 · Max: 5 · Default: 1.15 |
nag_alpha | number | Optional | The alpha value for NAG. This value is used as a final weighting factor for steering the normalized guidance (positive and negative prompts) in the direction of the positive prompt. Higher values will result in less steering on the normalized guidance where lower values will result in considering the positive prompt guidance more. · Max: 1 · Default: 0.25 |
nag_scale | number | Optional | The scale for NAG. Higher values will result in a image that is more distant to the negative prompt. · Max: 10 · Default: 3 |
base_shift | number | Optional | Base shift for the scheduled timesteps · Min: 0.01 · Max: 5 · Default: 0.5 |
fill_image | string | Optional | Use an image input to influence the generation. Can be used to fill images in masked areas. |
ip_adapters | object[] | Optional | IP-Adapter to use for image generation. · Default: [] |
aspect_ratio | string | Optional | The aspect ratio of the generated image. · Options: 21:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16 21:916:93:24:35:41:14:53:42:39:16 |
easycontrols | object[] | Optional | EasyControl Inputs to use for image generation. · Default: [] |
use_real_cfg | boolean | Optional | Uses classical CFG as in SD1.5, SDXL, etc. Increases generation times and price when set to be true. If using XLabs IP-Adapter v1, this will be turned on!. · Default: false |
control_loras | object[] | Optional | The LoRAs to use for the image generation which use a control image. You can use any number of LoRAs and they will be merged together to generate the final image. · Default: [] |
output_format | string | Optional | The format of the generated image. · Options: jpeg, png · Default: "png" jpegpng |
reference_end | number | Optional | The percentage of the total timesteps when the reference guidance is to be ended. · Min: 0 · Max: 1 · Default: 1 |
guidance_scale | number | Optional | Min: 0 · Max: 20 · Default: 3.5 |
real_cfg_scale | number | Optional | The CFG (Classifier Free Guidance) scale is a measure of how close you want the model to stick to your prompt when looking for a related image to show you. · Min: 0 · Max: 5 · Default: 3.5 |
sigma_schedule | string | Optional | Sigmas schedule for the denoising process. |
reference_start | number | Optional | The percentage of the total timesteps when the reference guidance is to bestarted. · Min: 0 · Max: 1 · Default: 0 |
use_beta_schedule | boolean | Optional | Specifies whether beta sigmas ought to be used. · Default: false |
reference_strength | number | Optional | Strength of reference_only generation. Only used if a reference image is provided. · Min: -3 · Max: 3 · Default: 0.65 |
num_inference_steps | integer | Optional | Min: 1 · Max: 50 · Default: 28 |
reference_image_url | string | Optional | URL of Image for Reference-Only |
change_map_image_url | string | Optional | URL of change map. |
02
Output Schema
| Field | Type | Description |
|---|---|---|
id | string | Unique identifier for the generation task |
status | string | Task status: pending, running, completed, failed, timeout |
model | string | Model used for the generation |
outputs | array | Array of output items |
outputs[].url | string | URL of the generated artifact |
outputs[].content_type | string | MIME type (e.g. image/png, video/mp4) |
error | object | null | Error details if failed, null on success |
error.type | string | Machine-readable error type code |
error.message | string | Human-readable error description |
Async Workflow
This model uses asynchronous execution. Submit a request and poll for the result.
- Submit — POST to /v1/run, receive an
id - Poll — GET /v1/run/{id} until status is
completed,failed, ortimeout - Retrieve — Read
outputsfrom the completed response
03
Code Examples
Ready-to-run snippets
# Step 1: Submit
curl -X POST https://api.sandbase.ai/v1/run \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "bfl/flux-general/differential-diffusion",
"image": "https://static.sandbase.ai/examples/bfl/flux-general/differential-diffusion/input_image_1.jpeg",
"loras": [],
"prompt": "Tree of life under the sea, ethereal, glittering, lens flares, cinematic lighting, artwork by Anna Dittmann & Carne Griffiths, 8k, unreal engine 5, hightly detailed, intricate detailed.",
"nag_end": 0.25,
"nag_tau": 2.5,
"strength": 0.85,
"max_shift": 1.15,
"nag_alpha": 0.25,
"nag_scale": 3,
"base_shift": 0.5,
"ip_adapters": [],
"easycontrols": [],
"use_real_cfg": false,
"control_loras": [],
"output_format": "png",
"reference_end": 1,
"guidance_scale": 3.5,
"real_cfg_scale": 3.5,
"reference_start": 0,
"use_beta_schedule": false,
"reference_strength": 0.65,
"num_inference_steps": 28,
"change_map_image_url": "https://static.sandbase.ai/examples/bfl/flux-general/differential-diffusion/input_change_map_image_url_0.jpeg"
}'
# Step 2: Poll result (replace <id>)
curl https://api.sandbase.ai/v1/run/<id> \
-H "Authorization: Bearer YOUR_API_KEY"API README
Flux General Differential Diffusion
Flux General Differential Diffusion belongs at the graded transformation inside a general editing workbench stage of a visual workflow. Rather than treating the model as an isolated demonstration, teams can place it inside briefing, review, selection, and handoff practices where the project can combine broad experimentation with careful control over how strongly different areas are reconsidered. This positioning clarifies why the model earns a place in a real creative pipeline and what kind of decision it helps people make.
A practical use of Flux General Differential Diffusion is selective stylization, nuanced restoration, partial redesign, and image studies where transitions between old and new should feel intentional. Begin by agreeing on the creative objective and review criteria, prepare only the source material needed for that objective, and compare results against audience, brand, editorial, and production needs. Technical request choices remain documented below so the prose can stay focused on planning and creative value.
Highlights
Spatially varying edit strength. Allows strong regeneration in selected areas while other regions change only subtly.
Smooth transition zones. Blends different denoising intensities without hard visual seams between regions.
Fine-grained creative control. Supports nuanced retouching, emphasis, and localized restyling beyond a binary mask.
Context-preserving synthesis. Uses surrounding image information to keep altered regions visually integrated. The General configuration exposes broad FLUX synthesis behavior for workflows that need explicit control over the selected transformation method.
Pricing
| Configuration | Billing unit | Price |
|---|---|---|
| Base generation | Per request | $0.00125 |
When to Use
| ✅ Good fit | ❌ Consider alternatives |
|---|---|
| The project needs this exact FLUX capability | The intended task belongs to a different media workflow |
| Required source and control media are available | Necessary assets or usage rights are unavailable |
| Creative direction can state change and preservation goals | Output must be deterministic at pixel or frame level |
| Supported dimensions and formats match final delivery | Placement requires unsupported specifications |
| An asynchronous generated result fits production | A live frame-synchronous response is mandatory |
Prompt Guide
Lead with the main subject and action, then specify composition, context, lighting, materials, style, typography, and atmosphere. For editing, identify each source and clearly separate the requested transformation from the subjects, regions, geometry, or identity that must remain unchanged.
{
"image": "https://example.com/start-frame.png",
"loras": [],
"nag_end": 0.25,
"prompt": "A cinematic, precisely composed result with a clear subject, controlled camera or viewpoint, realistic lighting, exact materials, and intentional atmosphere",
"seed": 1
}
Technical Specs
| Spec | Value |
|---|---|
| Model ID | bfl/flux-general/differential-diffusion |
| Input fields | seed (integer)<br>image (string)<br>loras (array)<br>prompt (string)<br>nag_end (number; …–1)<br>nag_tau (number)<br>strength (number; 0.01–1)<br>max_shift (number; 0.01–5)<br>nag_alpha (number; …–1)<br>nag_scale (number; …–10)<br>base_shift (number; 0.01–5)<br>fill_image (string)<br>ip_adapters (array)<br>aspect_ratio (string; 21:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16)<br>easycontrols (array)<br>use_real_cfg (boolean)<br>control_loras (array)<br>output_format (string; jpeg, png)<br>reference_end (number; 0–1)<br>guidance_scale (number; 0–20)<br>real_cfg_scale (number; 0–5)<br>sigma_schedule (string)<br>reference_start (number; 0–1)<br>use_beta_schedule (boolean)<br>reference_strength (number; -3–3)<br>num_inference_steps (integer; 1–50)<br>reference_image_url (string)<br>change_map_image_url (string) |
| Required input | prompt, image |
| Output fields | url, content_type |
| Execution | Asynchronous job |
Related Models
bfl/flux-general— Compare this concrete local family route.bfl/flux-general/image-to-image— Compare this concrete local family route.bfl/flux-general/inpainting— Compare this concrete local family route.