parameters

Aspect ratio of the output. Accepts `w:h` or one of the named aliases.

default

model-specific (usually 1:1)

values

1:12:33:24:55:49:1616:921:9squareportraitlandscapewideultrawidevertical

examples

cinematic portrait --ar 16:9phone wallpaper --ar 9:16poster art --ar 2:3

How many variations to return. Type the number directly — `--1`, `--2`, `--3`, or `--4`.

default

model-specific (usually 4 for images, 1 for videos)

values

--1--2--3--4

examples

a cyberpunk dragon --2quick sketch --1

Freeze the RNG. Same seed + same prompt + same model = same image. Some models may ignore it.

range

0 – 4294967295

examples

forest at dusk --seed 42portrait --seed 99999

Hard filter: only models that produce a native transparent (alpha) result are considered. Combines with model triggers; if the intersection is empty you get a clear error.

examples

app icon on transparent background --transparent

Style reference. Pass one or more image URLs; each can carry a per-URL weight via `::N`.

range

weight per URL: 0 – 1000

examples

a knight --sref https://cdn.example.com/style.jpgancient temple --sref https://a.jpg::500 https://b.jpg::200

Global style-reference weight. Applies to whatever `--sref` URLs are set.

range

0 – 1000

examples

portrait --sref https://x.jpg --sw 400

Character reference URL. Reserved — the parser accepts it but explicitly errors "not yet implemented" so it never silently no-ops.

note

Planned. Timeline unspecified.

Character-reference weight. Reserved alongside `--cref`.

range

0 – 100

note

Planned alongside `--cref`.

Force PNG output.

examples

icon --png

Force JPEG output.

examples

photo --jpg

Force WebP output. Falls back to PNG on providers that can’t emit WebP.

Force vector SVG output. Only picked when a vector-capable model can service the prompt.

examples

logo mark --svg

High-resolution image tier (~2048 px on the long side). Costs more credits.

examples

product hero shot --ar 16:9 --2k

Ultra-high-resolution image tier (~4096 px on the long side). Costs more credits.

examples

cinematic still --ar 16:9 --4k

Negative prompt — the things you don't want. Only a few of our models honor it as a hard exclusion; on the rest it becomes a hint that biases the prompt enhancer. The engine does NOT narrow the model pool when `--no` is present, so which behavior you get depends on which model the selector picks.

examples

forest landscape --no watermark, text, low-quality --ar 16:9

Upscale an input image (the `upscale_2x` operation).

examples

--upscale

Create a variation of an image (the `vary_strong` operation).

Subtle variation of an image.

Strong variation of an image.

Extend the canvas to the left. Append a companion `--<amount>` flag (fraction, e.g. `--0.5`, or pixels, e.g. `--512px`) to control how far.

range

amount fraction: 0.01 – 1.0 · pixels: 1 – 2048

examples

--panleft --0.5--panleft --512px

Extend the canvas to the right. Companion `--<amount>` flag controls how far.

range

amount fraction: 0.01 – 1.0 · pixels: 1 – 2048

examples

--panright --0.5--panright --512px

Extend the canvas upward. Companion `--<amount>` flag controls how far.

range

amount fraction: 0.01 – 1.0 · pixels: 1 – 2048

Extend the canvas downward. Companion `--<amount>` flag controls how far.

range

amount fraction: 0.01 – 1.0 · pixels: 1 – 2048

Zoom out 1.5×. Original stays in place; new content fills the border.

Zoom out 2×. Same idea as `--zoomout` but wider border.

note

Under development — the underlying `zoom_out_2x` operation is in the engine's DISABLED_OPERATIONS set; calls today will not run. Use `--zoomout` (1.5×) for now.

Zoom-out variant that expands the composition rather than just adding border.

note

Under development — the underlying `zoom_out_expand_1_5x` operation is in the engine's DISABLED_OPERATIONS set; calls today will not run.

Expand-out variant at 2×.

note

Under development — the underlying `zoom_out_expand_2x` operation is in the engine's DISABLED_OPERATIONS set; calls today will not run.

Trigger video generation. Becomes img2vid if input images are present, otherwise txt2vid.

examples

clouds rolling over mountains --mp4 --5sec --720p

Video resolution. Type the value directly — `--480p`, `--720p`, `--1080p`, etc. 4k is requested with `--2160p` or `--4k` (also `--4k60fps`) and is Seedance 2 Pro only — pair with `--seedance2pro` or `--flagship`. Note: `--4k` means the 4K *image* tier on image prompts, and 4k video on `--mp4` prompts. Frame-rate suffix (`p24`, `p30fps`) is also accepted; Seedance renders 24fps.

values

--480p--540p--720p--1024p--1080p--2160p / --4k (4k, Seedance 2 Pro)--480p24 / --480p24fps--540p24 / --540p24fps--720p24 / --720p24fps--1024p30 / --1024p30fps--1080p24 / --1080p24fps

examples

neon city --mp4 --1080pcinematic dragon --mp4 --seedance2pro --4k

Frame rate. 24 and 30 are the safest across providers; higher rates depend on the selected model.

values

--24fps--30fps--50fps--60fps

examples

drone shot --mp4 --24fps

Clip length. Accepts either the short `--5s` or the long `--5sec` form.

values

--4s / --4sec--5s / --5sec--6s / --6sec--8s / --8sec--10s / --10sec--12s / --12sec

examples

drone shot over reef --mp4 --10sec --1080p

Force the flagship (highest-quality) model of whatever operation you're running. Works universally.

examples

brand ad --ar 16:9 --flagshipcinematic clip --mp4 --flagship

Force OpenAI Sora 2 Pro (video).

forces

Sora 2 Pro

compat

Only one model trigger per prompt.

examples

glass shattering slow-mo --mp4 --sora2pro --1080p

Force OpenAI Sora Lite (video). Cheaper than Sora Pro.

forces

Sora Lite

Force Google Nano-Banana Pro. Works for both txt2img and img2img — the router picks the variant by input.

forces

Nano-Banana Pro

examples

product mockup --nanobananapro --2k

Force Google Nano-Banana 2 for txt2img / img2img.

forces

Nano-Banana 2

Force OpenAI GPT-image-2 (medium quality) for txt2img / img2img.

forces

GPT-image-2 (Medium)

Force GPT-image-2 (High quality). Slower + more expensive than `--gpt2`.

forces

GPT-image-2 (High)

Force Seedance 2 Fast (video). Works for txt2vid / img2vid / ref2vid.

forces

Seedance 2 Fast

Force Seedance 2 Pro (video). Works for txt2vid / img2vid / ref2vid.

forces

Seedance 2 Pro

Signed-in-only: access seeded demo generations without spending real credits.