GET

/v1/schemas/fal/image/fal-ai/lora/inpaint

Accept:

kind inputactivity imagehash 75b6ce34…extracted Nov 13, 2026version currentsource upstream spec ↗

field	recorded source
schema	Source linked above. upstream-spec Source trace `cabd4fbee147a95e63b801cc6452801ab0ad5397c9f29cb050648dde589c3cd8` { "fetchedAt": 1794546001, "extractorVersion": "2", "verifiedAt": null }

Request body

39 properties · 2 required · 6 $defs

property	type	description	constraints
model_name*	string	URL or HuggingFace ID of the base model to generate the image.
prompt*	string	The prompt to use for generating the image. Be as descriptive as possible for best results.
negative_prompt	string	The negative prompt to use.Use it to address details that you don't want in the image. This could be colors, objects, scenery and even the small details (e.g. moustache, blurry, low resolution).	default: ""
tile_height	integer	The size of the tiles to be used for the image generation.	128 ≤ n ≤ 4096 · default: 4096
timesteps	TimestepsInput	Optionally override the timesteps to use for the denoising process. Only works with schedulers which support the `timesteps` argument in their `set_timesteps` method. Defaults to not overriding, in which case the scheduler automatically sets the timesteps base…
unet_name	string	URL or HuggingFace ID of the custom U-Net model to use for the image generation.	nullable
image_encoder_weight_name	string	The weight name of the image encoder model to use for the image generation.	nullable
ic_light_model_background_image_url	string	The URL of the IC Light model background image to use for the image generation. Make sure to use a background compatible with the model.	nullable
num_images	integer	Number of images to generate in one request. Note that the higher the batch size, the longer it will take to generate the images.	1 ≤ n ≤ 8 · default: 1
tile_width	integer	The size of the tiles to be used for the image generation.	128 ≤ n ≤ 4096 · default: 4096
image_format	enum	The format of the generated image.	jpeg \| png · default: "png"
variant	string	The variant of the model to use for huggingface models, e.g. 'fp16'.	nullable
ic_light_model_url	string	The URL of the IC Light model to use for the image generation.	nullable
num_inference_steps	integer	Increasing the amount of steps tells Stable Diffusion that it should take more steps to generate your final result which can increase the amount of detail in your image.	1 ≤ n ≤ 150 · default: 30
embeddings	Embedding[]	The embeddings to use for the image generation. Only a single embedding is supported at the moment. The embeddings will be used to map the tokens in the prompt to the embedding weights.	items ≤ 1 · default: []
debug_latents	boolean	If set to true, the latents will be saved for debugging.	default: false
scheduler	enum	Scheduler / sampler to use for the image denoising process.	DPM++ 2M \| DPM++ 2M Karras \| DPM++ 2M SDE \| DPM++ 2M SDE Karras \| Euler \| Euler A · +5 more · nullable
noise_strength	number	The amount of noise to add to noise image for image. Only used if the image_url is provided. 1.0 is complete noise and values near 0 are minimal noise.	0 < n ≤ 1 · default: 0.5
ip_adapter	IPAdapter[]	IP adapters for image generation (up to 5). A non-empty list requires image_encoder_path.	items ≤ 5 · default: []
image_url	string	URL of image to use for image to image/inpainting.	nullable
ic_light_image_url	string	The URL of the IC Light model image to use for the image generation.	nullable
controlnets	ControlNet[]	The control nets to use for the image generation. You can use up to 5 control nets and they will be applied to the image at the specified timesteps.	items ≤ 5 · default: []
guidance_scale	number	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.	0 ≤ n ≤ 20 · default: 7.5
seed	integer	The same seed and the same prompt given to the same version of Stable Diffusion will output the same image every time.	nullable
loras	LoraWeight[]	The LoRAs to use for the image generation. You can use up to 10 LoRAs and they will be merged together to generate the final image.	items ≤ 10 · default: []
rescale_betas_snr_zero	boolean	Whether to set the rescale_betas_snr_zero option or not for the sampler	default: false
tile_stride_height	integer	The stride of the tiles to be used for the image generation.	64 ≤ n ≤ 2048 · default: 2048
image_encoder_subfolder	string	Subfolder containing the image encoder weights in image_encoder_path.	nullable
enable_safety_checker	boolean	If set to true, the safety checker will be enabled.	default: false
prompt_weighting	boolean	If set to true, the prompt weighting syntax will be used. Additionally, this will lift the 77 token limit by averaging embeddings.	default: false
controlnet_guess_mode	boolean	If set to true, the controlnet will be applied to only the conditional predictions.	default: false
tile_stride_width	integer	The stride of the tiles to be used for the image generation.	64 ≤ n ≤ 2048 · default: 2048
eta	number	The eta value to be used for the image generation.	0 ≤ n ≤ 1 · default: 0
mask_url	string	URL of black-and-white image to use as mask during inpainting.	nullable
image_encoder_path	string	Hugging Face ID or HTTPS URL of the image encoder; required when ip_adapter is non-empty. Choose an encoder compatible with the IP-Adapter weights. For h94/IP-Adapter SDXL weights, use h94/IP-Adapter with image_encoder_subfolder='sdxl_models/image_encoder'; for SD1.5 weights, use image_encoder_subfolder='models/image_encoder'.	nullable
prediction_type	enum	The type of prediction to use for the image generation. The `epsilon` is the default.	v_prediction \| epsilon · default: "epsilon"
clip_skip	integer	Skips part of the image generation process, leading to slightly different results. This means the image renders faster, too.	0 ≤ n ≤ 2 · default: 0
sigmas	SigmasInput	Optionally override the sigmas to use for the denoising process. Only works with schedulers which support the `sigmas` argument in their `set_sigmas` method. Defaults to not overriding, in which case the scheduler automatically sets the sigmas based on the `nu…
debug_per_pass_latents	boolean	If set to true, the latents will be saved for debugging per pass.	default: false

Types

6 $defs

ControlNetobject

config_url	string	optional URL to the controlnet config.json file.	nullable
mask_url	string	The mask to use for the controlnet. When using a mask, the control image size and the mask size must be the same and divisible by 32.	nullable
image_url*	string	URL of the image to be used as the control net.
path*	string	URL or the path to the control net weights.
start_percentage	number	The percentage of the image to start applying the controlnet in terms of the total timesteps.	0 ≤ n ≤ 1 · default: 0
conditioning_scale	number	The scale of the control net weight. This is used to scale the control net weight before merging it with the base model.	0 ≤ n ≤ 2 · default: 1
end_percentage	number	The percentage of the image to end applying the controlnet in terms of the total timesteps.	0 ≤ n ≤ 1 · default: 1
ip_adapter_index	integer	The index of the IP adapter to be applied to the controlnet. This is only needed for InstantID ControlNets.	nullable
variant	string	The optional variant if a Hugging Face repo key is used.	nullable

Embeddingobject

tokens	string[]	The tokens to map the embedding weights to. Use these tokens in your prompts. When omitted, tokens are generated as <s0>, <s1>, ... to match the number of vectors in the embedding file.	nullable
path*	string	URL or the path to the embedding weights.

IPAdapterobject

image_projection_shortcut	boolean	The value to set the image projection shortcut to. For FaceID plus V1 models, this should be set to False. For FaceID plus V2 models, this should be set to True. Default is True.	default: true
insight_face_model_path	string	URL or the path to the InsightFace model weights.	nullable
ip_adapter_mask_url	string	The mask to use for the IP adapter. When using a mask, the ip-adapter image size and the mask size must be the same	nullable
model_subfolder	string	Subfolder in the model directory where the IP adapter weights are stored.	nullable
unconditional_noising_factor	number	The factor to apply to the unconditional noising of the IP adapter.	0 ≤ n ≤ 1 · default: 0
scale	number	The scale of the IP adapter weight. This is used to scale the IP adapter weight before merging it with the base model.	n ≥ 0 · default: 1
path*	string	URL or the path to the IP adapter weights.
scale_json	object	The scale of the IP adapter weight. This is used to scale the IP adapter weight before merging it with the base model.	nullable
ip_adapter_image_url*	string \| string[]	URL of the image to be used as the IP adapter, or a list of up to 5 URLs.
weight_name	string	Name of the weight file.	nullable

LoraWeightobject

scale	number	The scale of the LoRA weight. This is used to scale the LoRA weight before merging it with the base model.	0 ≤ n ≤ 4 · default: 1
path*	string	URL or the path to the LoRA weights.

SigmasInputobject

method	enum	The method to use for the sigmas. If set to 'custom', the sigmas will be set based on the provided sigmas schedule in the `array` field. Defaults to 'default' which means the scheduler will use the sigmas of the scheduler.	default \| array · default: "default"
array	number[]	Sigmas schedule to be used if 'custom' method is selected.	default: []

TimestepsInputobject

method	enum	The method to use for the timesteps. If set to 'array', the timesteps will be set based on the provided timesteps schedule in the `array` field. Defaults to 'default' which means the scheduler will use the `num_inference_steps` parameter.	default \| array · default: "default"
array	integer[]	Timesteps schedule to be used if 'custom' method is selected.	default: []

Validate a payload

before you spend tokens

shell

$ curl -X POST https://modelschemas.com/v1/validate \
    -d '{"provider":"fal","endpointId":"fal-ai/lora/inpaint","payload":{…}}'
→ {"valid": false, "errors": [{"path": "#", "keyword": "required", …}]}