qwen/qwen-image

An image generation foundation model in the Qwen series that achieves significant advances in complex text rendering.

Capabilities

Reference ImagesNegative PromptSeed

Cost

Community model (estimated from hardware time)

Input Parameters

Name	Type	Description	Default	Constraints
`prompt`*	string	Prompt for generated image	`—`	—
`aspect_ratio`	string	Aspect ratio for the generated image	`"16:9"`	1:116:99:164:33:43:22:3
`disable_safety_checker`	boolean	Disable safety checker for generated images.	`false`	—
`enhance_prompt`	boolean	Enhance the prompt with positive magic.	`false`	—
`extra_lora_scale`	array	Scales for additional LoRAs as an array of numbers (e.g., 0.5, 0.7). Must match the number of weights in extra_lora_weights.	`—`	—
`extra_lora_weights`	array	Additional LoRA weights as an array of URLs. Same formats supported as lora_weights (e.g., ['https://huggingface.co/flymy-ai/qwen-image-lora/resolve/main/pytorch_lora_weights.safetensors', 'https://huggingface.co/flymy-ai/qwen-image-realism-lora/resolve/main/flymy_realism.safetensors'])	`—`	—
`go_fast`	boolean	Run faster predictions with additional optimizations.	`true`	—
`guidance`	number	Guidance for generated image. Lower values can give more realistic images. Good values to try are 2, 2.5, 3 and 3.5	`3`	min: 0, max: 10
`image`	string(uri)	Input image for img2img pipeline	`—`	—
`image_size`	string	Image size for the generated image	`"optimize_for_quality"`	optimize_for_qualityoptimize_for_speed
`lora_scale`	number	Determines how strongly the main LoRA should be applied.	`1`	—
`lora_weights`	string	Load LoRA weights. Only works with text to image pipeline. Supports arbitrary .safetensors URLs, tar files, and zip files from the Internet (for example, 'https://huggingface.co/flymy-ai/qwen-image-lora/resolve/main/pytorch_lora_weights.safetensors', 'https://example.com/lora_weights.tar.gz', or 'https://example.com/lora_weights.zip')	`—`	—
`negative_prompt`	string	Negative prompt for generated image	`" "`	—
`num_inference_steps`	integer	Number of denoising steps. Recommended range is 28-50, and lower number of steps produce lower quality outputs, faster.	`30`	min: 1, max: 50
`output_format`	string	Format of the output images	`"webp"`	webpjpgpng
`output_quality`	integer	Quality when saving the output images, from 0 to 100. 100 is best quality, 0 is lowest quality. Not relevant for .png outputs	`80`	min: 0, max: 100
`replicate_weights`	string	Load LoRA weights from Replicate training. Only works with text to image pipeline. Supports arbitrary .safetensors URLs, tar files, and zip files from the Internet.	`—`	—
`seed`	integer	Random seed. Set for reproducible generation	`—`	—
`strength`	number	Strength for img2img pipeline	`0.9`	min: 0, max: 1

promptrequiredstring

Prompt for generated image

aspect_ratiostring

Aspect ratio for the generated image

Default: "16:9"

1:116:99:164:33:43:22:3

disable_safety_checkerboolean

Disable safety checker for generated images.

Default: false

enhance_promptboolean

Enhance the prompt with positive magic.

Default: false

extra_lora_scalearray

Scales for additional LoRAs as an array of numbers (e.g., 0.5, 0.7). Must match the number of weights in extra_lora_weights.

extra_lora_weightsarray

Additional LoRA weights as an array of URLs. Same formats supported as lora_weights (e.g., ['https://huggingface.co/flymy-ai/qwen-image-lora/resolve/main/pytorch_lora_weights.safetensors', 'https://huggingface.co/flymy-ai/qwen-image-realism-lora/resolve/main/flymy_realism.safetensors'])

go_fastboolean

Run faster predictions with additional optimizations.

Default: true

guidancenumber

Guidance for generated image. Lower values can give more realistic images. Good values to try are 2, 2.5, 3 and 3.5

Default: 3min: 0, max: 10

imagestring

Input image for img2img pipeline

image_sizestring

Image size for the generated image

Default: "optimize_for_quality"

optimize_for_qualityoptimize_for_speed

lora_scalenumber

Determines how strongly the main LoRA should be applied.

Default: 1

lora_weightsstring

Load LoRA weights. Only works with text to image pipeline. Supports arbitrary .safetensors URLs, tar files, and zip files from the Internet (for example, 'https://huggingface.co/flymy-ai/qwen-image-lora/resolve/main/pytorch_lora_weights.safetensors', 'https://example.com/lora_weights.tar.gz', or 'https://example.com/lora_weights.zip')

negative_promptstring

Negative prompt for generated image

Default: " "

num_inference_stepsinteger

Number of denoising steps. Recommended range is 28-50, and lower number of steps produce lower quality outputs, faster.

Default: 30min: 1, max: 50

output_formatstring

Format of the output images

Default: "webp"

webpjpgpng

output_qualityinteger

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

Default: 80min: 0, max: 100

replicate_weightsstring

Load LoRA weights from Replicate training. Only works with text to image pipeline. Supports arbitrary .safetensors URLs, tar files, and zip files from the Internet.

seedinteger

Random seed. Set for reproducible generation

strengthnumber

Strength for img2img pipeline

Default: 0.9min: 0, max: 1

Version: 0bba9e70f784Updated: 7/25/20261.6M runs